What determines Financial Depth and Intermediation Efficiency of banks in Sub-Saharan Africa?

(1)

What determines Financial Depth and Intermediation

Efficiency of banks in Sub-Saharan Africa?

A two-step approach

Master Thesis Abstract

This study aims to find out why banks in Sub-Saharan Africa (SSA) invest only a small portion of their deposits in loans to the private sector but rather invest in liquid assets and therefore exhibit excess liquidity compared to other developing economies. The prerequisite for an efficient intermediation of banks is a large volume of credit relatively to GDP, which indicates the level of financial depth and is one of the financial development measures. Based on that fact, this study takes a two-step approach to explain what determines the intermediation efficiency of banks. In the first step, the study aims to find out the determinants of the volume of credit in the financial systems of SSA measured by private credit to GDP. In the second step, the factors determining the intermediation efficiency of banks are identified, which is measured by bank credit to total deposits. The empirical findings show that the legal structure (regulatory quality) and the ownership structure (foreign banks presence) significantly influence financial depth. Furthermore, government bonds crowd-out credit to the private sector. Moreover, an additional interaction model indicates that the impact of the determinants differ across income levels. The results are consistent over several robustness checks, amongst others utilizing the instrumental variable two-stage least square approach.

Key Words: Financial development, Financial depth, Intermediation efficiency, Sub-Saharan

Africa

Annette Juma Boliba (s2752433) a.juma.boliba@student.rug.nl

(2)

List of Figures

Figure 1: Banks’ balance sheets in developing country and developed country ... 1!

Figure 2: Comparison of financial development between African and non-African countries ... 5!

Figure 3:Mean of financial depth measures in Sub-Saharan Africa ... 6!

Figure 4: Mean of intermediation efficiency of banks in Sub-Saharan Africa by income levels ... 14!

Figure 5: Mean of financial depth in Sub-Saharan Africa by income levels ... 15!

Figure 6: Cook’s D of key dependent variables ... 19!

List of Tables

Table 1: Descriptive Statistics ... 13!

Table 2: Fixed effects Model and IV 2SLS estimation of Determinants of Private credit to GDP .. 23!

Table 3: Fixed effects Model and IV 2SLS estimation of Determinants of Bank credit to total deposits ... 25!

Table 4: Interaction Model of Private credit to GDP with income level dummies ... 28!

Table 5: Interaction Model of Bank credit to total deposits with income level dummies ... 29!

List of Abbreviations

2SGLS Two-stage generalized least-squares

2SLS Two-stage least-squares

ADI African Development Indicators

FDSD Financial development & structure dataset GFDD Global Financial Development Databank

GMM Generalized method of moments

IMF International Monetary Fund

IV Instrumental variables

LDC Least developed countries

SSA Sub-Saharan Africa

VIF Variance inflation factor

WDI World Development Indicators

WEO World Economic Outlook

(4)

1. Introduction

It has often been observed that banks in developing countries tend to be investment constrained (Freedman & Click, 2006). This is revealed through a look on a banks’ balance sheets. Banks in developing countries invest a great part of their deposits in liquid assets such as government bonds rather than offering loans to the private sector. The self-constructed example in figure 1 illustrates the behaviour, where banks having a large deposit base prefer to utilize it for liquid assets rather than providing loans. This phenomenon is of importance since economic theory, as well as empirical analysis claim that holding liquid assets has a limited link to economic growth compared to loans to the private sector. These loans can help to build up real capital and be used for productive projects in order to foster growth. A dysfunctional credit market is characterised by inefficiencies of banks to allocate credit productively. These inefficiencies need to be analysed precisely in order to understand the low level of bank lending to the private sector. Especially African countries exhibit this credit constraint behaviour, which is further discussed in the following sections.

This paper will proceed as follows: Based on theory, the role of the financial system and the banks’ functions as intermediaries will be explained. Thereafter features of financial development and their relationship to growth will be presented, followed by an overview of the situation of banks in Africa in comparison to other non-African countries. Based on that, a model is constructed to determine the factors impacting financial depth and the intermediation efficiency of banks in Sub-Saharan African. Moreover, the study will closely examine if the effects differ depending on the income levels.

Figure 1: Banks’ balance sheets in developing country and developed country

Notes: The figure is a self-constructed example and illustrates a bank’s balance sheet in developing country (left) and developed country (right). The numbers are chosen to show the credit constraint in developing countries, which as depicted here will invest only a small portion of deposits in loans, whereas the bank in a developed country performs the opposite. (Freedman & Click, 2006)

2. Literature review

(5)

financial system. Moreover, it gathers empirical literature exploring the link between financial development and economic growth. In this respect financial development in African countries and the overall structure of the banking system is summarized in order to explain the bank lending behaviour to the private sector of banks in Sub-Saharan Africa.

2.1. Role of financial System and Banks as Intermediaries

The theory gives a clear definition of the role of a financial system. It is defined as “a set of markets for financial instruments, and the individuals and institutions, who trade in those markets” (Howells & Bain, 2008). Among other things, the financial system is supposed to channel funds from lenders (surplus units) to borrowers (deficit units). One of the approaches to ensure that funds are channelled from the one who is in surplus to another one who is in need, is to use banks as intermediaries. Banks’ role as intermediaries is to facilitate lending and borrowing by making both cheaper and therefore stimulate investment and spending. However, this is assumed to play out in a developed financial system rather than in a less developed financial system. This is expected to lead to poor growth performance of the latter economies. In particular, banks transform funds to loans, which are provided to borrowers (Howells & Bain, 2008), (Kohn, 2004), (Bertocco, 2009), (Levine, 2004). Financial institutions such as banks are expected to reduce transaction costs by bringing together lender and borrower more quickly and efficiently. Furthermore financial institutions are assumed to solve informational asymmetries, so that both sides, the bank and the depositor, have an increased incentive to engage in loan contracts. Thus aggregate investment, as well as the quality of loans, is expected to increase, especially because banks intend to maximize their holdings of investments and loans and to minimize their reserves (Bhattacharya, 1993). Theory suggests that loans and deposits increase simultaneously, since “for every loan created, someone must receive an addition to his or her deposit” (Howells & Bain, 2008).

(6)

for the economy. They are considered as a central authority determining where the most productive use of resources is located. If they fail to fulfil this task, money might be allocated to less productive projects and this in turn might deteriorate economic growth (Schumpeter, 1939, [1964]), (Bertocco, 2009) (Dullien, 2009). This is especially the case, if banks do not allocate a large portion of credit where the highest profit is expected.

2.2. Financial development and economic growth

There are plenty of studies investigating the link between financial development and economic growth. Many arrive to the conclusion that financial development is positively associated with economic growth. However, their methods and explanations differ in this matter.

Demetriades and James (2011) empirically investigate the relationship between financial development and economic growth using a panel estimation method on 18 Sub-Saharan African countries. They conclude that the link between both factors in SSA is “broken” (Demetriades and James 2011, p. 9). Financial development is measured with the variables bank deposits to GDP, liquid liabilities to GDP and private credit by deposit money banks to GDP. Each variable is regressed on economic development, reflected by real GDP per capita. In contrast to previous research estimating the relation between financial development and economic growth finding a positive or a negative correlation between both measures, Demetriades and James (2011) do not find any link between the common measure of financial development bank credit and economic growth (King & Levine, 1993), (Levine, 2004). Nevertheless, their panel co-integrations estimation approach shows a positive and significant correlation between the dependent variable liquid liabilities and real GDP per capita. The results emphasize that financial development as such does not necessarily have an impact on the real sector, since no connection between bank credit and real GDP per capita has been estimated.

(7)

King and Levine (1993) use four indicators for financial development in order to evaluate financial depth utilising data on 80 countries over the years 1960 to 1989 in a cross-country estimation approach. Two of those indicators measure domestic asset distribution to the private sector. The “credit issued to non-financial private divided by total credit” (excluding credit to banks) is that “financial systems that primarily fund private firms probably provide more services than financial systems that simply funnel credit to the government or state enterprises”. Expecting that firms have diverse needs from the financial sector, they argue that a sector that can meet these needs is probably more efficient in producing information and allocating capital because it is much harder to find the most productive firms in an economy than it is to simply give the capital to the government. Hence, the higher the share of loans to the private sector the more efficient the financial system (King & Levine, 1993).

A negative relationship between financial development and economic growth has been detected recently. In particular, credit stock a common indicator of financial development is not significantly positively correlated to growth, even after controlling for positive effects deriving from credit flows (annual change of credit stocks relative to lagged GDP). Hence, financial deepening does not cause positive growth. This applies especially when credit stocks support asset markets rather than productive investments and innovation (Bezemer, Grydaki, & Zhang, 2014). These findings are in line with those of Bezemer (2013) who establishes a crucial difference between investment in financial capital and real capital. Investment in real capital refers to the use of capital in productive projects consisting of fixed capital such as machineries. On the other hand, investment in financial capital implies investment in financial assets such as bonds and portfolios. The distinction is made to figure out what factors lead to economic growth. Based on data extracted from the WDI and incorporating multiple LDCs, Bezemer (2013) finds evidences that an increase in financial capital has not led to economic growth. According to him money has been used for financial assets and not to build up real capital such as productive projects, which are assumed to enhance economic growth. In other words, lending to the private sector, which can be mobilized for productive investments, fosters growth.

2.3. Financial system in Africa in international comparison

It is generally known that the financial systems of several African countries do not perform as well as those in other regions. This section intends to illustrate differences between the financial sectors in African countries in comparison to those in other parts of the world and to other developing economies.

(8)

she concludes that the level of financial development often expressed through measures of financial depth and efficiency, differs around the world and in particular across developing countries with similar income level. Due to data restrictions she relies on common measures to explain cross-country differences with respect to financial development. The analysis shows that private credit and market capitalization rises with income. While Soudan and Ethiopia, both having a low GDP per capita, experience and a low degree of private credit to GDP, advanced economies such as the USA have a large private credit to GDP per capita ratio.

The current status of Africa’s banking system is summarized by Beck and Cull (2013). According to them, it has experienced considerable changes over the last years. In their analysis, Africa’s banking system is compared to other countries with similar income level. Thereby it appears that within Africa, there is a high variation in the development of financial systems. To examine the financial system as such looking at the level of financial depth has proven to be enlightening. Financial depth is measured through the deposit based in the financial system and the credit provided through the system. Proxies are therefore liquid liabilities to GDP, which measures monetary resources accumulated by banks relative to economic activity, bank deposits to GDP and private credit to GDP, which gives an overview over the strength of financial intermediation relative to economic activity.

Figure 2: Comparison of financial development between African and non-African countries

Source: Beck and Cull (2013)

Notes: To arrive to a proper benchmark for African banking systems the authors only consider low-/lower middle income countries in Africa and compare the median to the median country across a sample of low-/lower middle income non-African countries

(9)

6

compared to 47% in other low- and lower-middle income countries outside Africa. The bank deposit ratio to GDP is 25% in Africa, while non-African countries account for 38%. The private credit to GDP is 18% in Africa compared to almost twice as much in non-African countries (34%) in 2011. All in all, African countries account for a lower degree of financial development than non-African developing countries. The difference between bank deposits to GDP and private credit illustrates the difficulties of African banks to allocate the savings and funds to credits to the private sector compared to other developing countries outside Africa.

To particularly look at Sub-Saharan Africa, the means of the different financial depth measures have been extracted respectively from the Financial Development and Structure dataset (FDSD) and the Global Financial Development Database of the World Bank (GFDD). Figure 3 illustrates the evolution of financial depth measures over the years 2004 until 2011 for SSA. Overall, an upward trend can be observed for all three measures of financial development. The mean of liquid liabilities to GDP has increased from 31% in 2004 to 39% in 2011. Bank deposits to GDP rises from 26% in 2004 to 34% in 2011, while the mean of Private credit to GDP increases from 19% to 26% for the same years. However, financial systems in SSA are still shallow when comparing them to those in non-African countries as shown in figure 2. The values in figure 3 differ from those calculated by Beck and Cull (2013) insofar that figure 3 includes omitted countries such as South Africa and Mauritius. The constant and linear gap between bank deposits to GDP and private credit to GDP suggests that financial institutions are reluctant to make use of their deposits to provide loans to the private sector.

Figure 3:Mean of financial depth measures in Sub-Saharan Africa

Note: Dataset contains 41 selected Sub-Saharan Africa countries. Source: FDSD (2013) & GFDD (2013)

Figure A in appendix 2 shows the financial depth indicators in Sub-Saharan Africa divided by income level. The figure illustrates that financial development differs across income levels.

0! 5! 10! 15! 20! 25! 30! 35! 40! 45! 2004! 2005! 2006! 2007! 2008! 2009! 2010! 2011!

Financial Depth in SSA (%), 2004-2011

(10)

Moreover, there is some indication that financial depth correlates with the income classification of countries. This is confirmed when looking at figure A. The higher the income level is the higher the depth of the financial system is. Low-income economies experience the lowest level of financial depth. Their mean for liquid liabilities is 31% while bank deposits to GDP is 24% and private credit to GDP rather low with 16%. The lower-middle-income economies exhibit a private credit ratio of 22% whereas the upper-middle-income economies have a private credit ratio of 40%. These findings underline the large variation of financial deepening in African countries and the fact that higher-income countries have a higher level of financial development (Oluitan, 2014), (Allen & Santomero, 1998), (Beck & Honohan, 2007). Beck and Honohan (2007) claim that these differences can be mostly explained by national per capita income, controlling for inflation. In line with this perception, upper-middle-income economies such as South African and Mauritius have a higher level of financial depth than low-income countries such as Benin or the Republic of Congo. For the purpose of this study, it is important to understand that African banking systems are shallower than non-African banking systems with respect to credit provided to the private sector. The limited intermediation shows up in the low level of loan to deposit ratio to the private sector in comparison to the high amount of liquid assets hold by banks. It has been empirically justified that the level of credit to the private sector contributes to economic growth. Therefore, fostering the intermediation process is critical for the financial development of banks in Sub-Saharan Africa. The rationale of the low level of credit to the private sector will be discussed further later in this study. Nevertheless, an improvement of all measures across all countries took place over the past years, which emphasizes the growth and development potential of financial systems in developing areas such as SSA (Beck & Honohan, 2007), (Beck & Cull, 2013).

(11)

accounts for 10.3% compared to 8.2% in banks outside Africa (Beck & Cull, 2013). Hence, banking in Africa is more expensive, which could be a potential factor of low lending rates. These features of banks in Africa are coupled with a low level of bank competition and high bank concentration. Although there has been a downward trend, the five-bank concentration ratio in 2009 still amounted to 77.29 %.1_{It indicates that the high level of bank concentration is consistent}

with the limited bank competition in the African region (Fosu, 2013).

The ownership structure of banks in Africa has developed in recent years, as quoted by several studies. The privatization process of banks in many African countries, such as Kenya, Uganda, Tanzania, Rwanda and Cameroon has encouraged growth in the number of banks. While in the past the ownership of banks was mostly granted to the government, recently a large part of banks is owned by foreign banks (Allen, Otchere, & Senbet, 2010), (Fosu, 2013). Foreign banks are assumed to be more profitable, more efficient than locally owned and especially state-owned banks, which would legitimize the privatization (Allen, Otchere, & Senbet, 2010), (Fosu, 2013), (Beck & Honohan, 2007), (Beck & Cull, 2013), (Andrianaivo & Yartey, 2009).

Overall, the banking systems in Sub-Saharan Africa have experienced growth in recent years. However, the growth is not consistent with persistent shallowness of the financial system, which is reflected amongst others through the low level of intermediation to the private sector. African banking systems are still less developed than those in other developing countries. Banks are over-liquid compared to other developing regions. Nevertheless, they invest a big portion of their deposit in liquid assets, such as government bonds, rather than providing the private sector with credit. The model in this paper aims to examine what factors influence the behaviour of banks in SSA, discussed in previous studies. In particular, the model investigates potential determinants of financial depth and intermediation efficiency of banks. The analysis is performed in section 3.

2.4. Bank lending to the private sector

Previous studies intended to explain the link between the level of credit to the private sector and economic development in developing countries, but just few attempted to find the determinants of the low level of credit to the private sector. This research paper is based on Freedman and Click (2006) who investigate the low level of bank lending in developing countries by focussing on the level of liquidity prevailing in the banking systems of those economies. They conclude that the low level of bank lending is a reflection of the large allocation of deposits in liquid assets, such as government bonds rather than increased lending to private sector. This supports the findings discussed in the previous section illustrating that banks in African countries are over-liquid but remain reluctant to provide loans to the private sector. Excess liquidity in the banking system of

1

(12)

developing countries is estimated by comparing their liquidity with the amount of liquidity that would prevail in the US for the same deposit base. Inefficiency of the banking system is accounted for by examining the impact of the amount of liquid liabilities and that of the level of credit respectively on economic growth. However, their reasoning with respect to the differences between the credit ratio and the liquidity ratio is based on previous findings and not on own empirical estimation results. According to them, the inefficiency of the credit market is influenced by various factors such as high reserve requirement, poor legal and regulatory regimes including contract enforcement issues, bankruptcy regime, imperfect collateral laws as well as fiscal policy issues, and asymmetric information. Moreover just a limited amount of African countries are taken into account. But as shown in figure 3, there is a wide variation in financial development within African countries, which has to be considered in order to arrive to meaningful conclusions.

Andrianova et al. (2011) approach the problem of low level of loans to the private sector theoretically and empirically. An IO-model of banking serves for the theoretical approach whereas the empirical part is derived through a Generalized Method of Moments (GMM) dynamic panel estimator model. According to them, predominant determinants of the low level of loans as a share of assets are moral hazard originating from strategic loan defaults and adverse selection arising from a lack of good projects. Using a sample of 378 African banks they deduce that loan defaults, reflected through the variable impaired loans to total loans, are the main significant factor hindering bank lending when the regulatory quality is under a certain threshold. Once this threshold is exceeded the impact of regulatory quality remains limited and insignificant on the volume of loans. They include country and bank fixed effects to control for differences between countries and banks. Furthermore, the inclusion of random error in the model accounts for irregular changes in loan volumes caused for instance by shocks. However, they do not observe bank specific factors such as liquidity ratio and loan to deposit ratio, which might also influence the lending decision of banks (Andrianova, Demetriades, & Fielding, 2010). Moreover, there is no distinction between loans granted to the private sector, which, as already explained, is suggested to contribute more to economic growth than loans given to the public sector for instance.

(13)

environment with a volatile macro-economic outcome, banks are more cautious and reluctant to lend, because of uncertainty with respect to profitable investment opportunities and returns. This results in a lower level of bank lending. His analysis of factors impacting bank lending is subdivided in three steps. First, cross-country determinants are analysed using a general least square estimation method and random effect estimator. Thereafter, he includes financial reform and banking freedom measures into the model. For robustness reasons, he focuses on regional determinants of bank credit. Therefore the GMM estimator is utilized. The empirical results emphasize that bank credit to the private sector is particularly influenced by bank-specific characteristic such as for instance excess liquidity2, the efficiency of the management, and bank growth. The regional subdivision reveals that determinants of bank credit delivery vary across regions. While excess liquidity is highly significantly negatively correlated to bank loans only in the West African States (ECOWAS) and the East African Community (EAC), management efficiency is positively significantly related only to the Southern African Development Community (SADC). However, Amidu (2014) defines bank credit to the private sector with the variable credit to the private sector as share of GDP. Yet, this variable depicts only the volume of credit to GDP, but does not give any information about the efficiency of banks in providing credit to the private sector.

Rashid (2011) empirically finds a negative and highly significant correlation between bank lending in developing countries and foreign bank presence. This research contradicts several studies showing a positive link between foreign banks presence and financial development in developing countries (Claessens, Demirguc-Kunt, & Huizinga, 1998). The proponents of foreign banks entry in less developed financial systems argue that foreign banks generally foster the financial structure through new and advanced technical skills in financial products and financial innovation. They integrate them in the hosting financial markets and hence reduce costs (Levine, 1996). However, Rashid (2011) claims that foreign banks have no incentive to engage in domestic bank lending. According to him, they already benefit from their comparative advantage to earn profits from activities other than lending. They engage for instance in security trading, and stock or bond markets. Thus they are less vulnerable to economic crisis, which makes it difficult to recollect loans from the private sector. Furthermore they suffer from information asymmetry about local borrowers making them reluctant to give out loans to the private sector. Additionally, the author offers an alternative explanation, which in my knowledge, has never been considered. His findings indicate that domestic banks lose their deposit base to foreign banks and thus have less back up to provide loans to the private sector. This is justified by a negative and significant correlation

(14)

between the deposit share of foreign banks and domestic credit to the private sector to GDP. Applying the GMM estimation technique on a sample covering bank-level data of 81 developing and emerging countries it appears that foreign banks lend less to the private sector. Thereby the dependent variable reflecting the share of loans as a fraction of a bank’s total assets shows a negative and significant correlation with foreign bank ownership. This holds even after controlling for various macroeconomic conditions. Another regression with a GMM estimation method demonstrates that the total share of deposits held in foreign banks is significantly and negatively correlated with the level of credit to the private sector. As in previous studies, Rashid (2011) controls for macroeconomic indicators such as regulatory environment, which according to the results influence banks’ lending behaviour.

In line with Rashid’s findings, Claessens and van Horen (2012) also find a negative relationship between foreign bank presence and domestic credit creation. Applying a cross-section model for the years 1995-2009 they find that the share of foreign bank assets to total assets has a negative relationship to credit to the private sector. However, when dividing their set of countries in OECD, emerging, and developing countries only the latter countries show a significant and negative correlation to private credit to GDP. They investigate the relationship between total lending of banks and the presence of foreign banks while leaving out the bank efficiency measure reflected through the ratio of bank credit to total deposits. Bank credit to total deposits seems to be more appropriate to examine how foreign banks presence influences the intermediation efficiency of banks.

While many studies find evidence for the shallowness financial system in Africa, Iossifov and Khamis (2009) focus on the growth rate of real bank credit to the private sector in SSA. Moreover, they concentrate on the development of the ratio of bank credit to the private sector to non-oil GDP. According to their 2SLS regression estimation results, an increase in per capita GDP is associated with a rise in the credit availability. A boost in the nominal interest rates is related to a reduction in the ratio of bank credit to the private sector. They also address the challenges associated with the larger credit availability. As a consequence, imprudent credit expansion can cause imbalances, higher external debt, and increased borrowing from the private sector can lead to an interest rates boost.

(15)

government bond issuance leads to a reduction in domestic investment and borrowing, especially if the financial system is weak. Hence, a 1% increase of government issuing debt reduces credit to the private sector by 0.30 US dollars.

This study contributes to the existing literature in several ways. It aims to show the determinants of financial depth and the intermediation efficiency of banks in SSA using both, the private credit by deposit money bank to GDP ratio and the bank credit to deposits ratio. The former reflects the strength of financial intermediation relative to economic activity. It gives information about the financial depth in a country and also shows how much credit is available. The latter shows how much bank deposits are invested in credit to the private sector. The higher the ratio, the lower the liquidity ratio3_{by definition and thus the higher the intermediation efficiency is expected to be. It}

has been highlighted that the environment in SSA substantially affects the level of credit to the private sector. This motivates the assumption that the intermediation efficiency of banks can be explained by the ownership structure of banks, the legal structure, and macroeconomic conditions prevailing in these countries. Additionally, the effects of the determinants on the dependent variables are investigated for different income levels to account for the high variation of financial development and intermediation efficiency of banks across Sub-Saharan African countries.

3. Data and Methodology 3.1. Descriptive Statistics

The sample consists of country level data on 28 countries in Sub-Saharan Africa over the time period from 2004 until 2011. The countries are classified in lower-income, lower-middle-income and upper-middle-lower-middle-income economies4 respectively to capture the heterogeneity of intermediation efficiency across SSA countries. Due to restricted data availability different databases have been used to construct the sample. Data on financial indicators is extracted from the Financial Development and Structure database (FDSD) and complemented by the Global Financial Development Database (GFDD), Bank Scope and the IMF World Economic Outlook (WEO). Bank level data is aggregated on a country level to allow for cross-country comparisons. Moreover, data on macroeconomic indicators is gathered from the Worldwide Governance indicators (WGI), African Development Indicators (ADI) and Fraser Institute’s Economic Freedom Indicators. Table 1 reports the descriptive statistics of the variable used for the model specification. The mean for the private credit ratio by deposit money bank to GDP5_{is 22% and the mean for bank credit to total}

deposits is 78%.

3_{Liquidity ratio is measured by the liquid assets to total deposits ratio (see GFDD, World Bank)}

4

The cut off between income-levels is adopted from the GFDD, World Bank

(16)

Table 1: Descriptive Statistics

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

VARIABLES Label N Mean Sd Max Min

Private_C Private credit to GDP 154 21.75 18.63 86.72 2.036

Bdep Bank deposits to GDP 154 27.14 18.23 92.33 6.108

Creddep Bank credit to total deposits 156 75.08 21.01 129.2 21.07 Liq_A Liquid assets to total deposits 153 37.57 14.58 75.91 10.43

RQ Regulatory quality 156 -0.417 0.520 0.898 -1.358

Foreign_A Foreign bank assets to total _assets

151 62.54 27.49 100 0

G_debt Government bonds 156 37.47 26.77 141.2 5.983

loginf Log (inflation rate) 153 1.807 0.917 3.793 -2.919

cmr Credit market regulations 151 8.023 1.315 10 4.400

GDP_growth GDP per capita growth 156 2.762 3.380 18.99 -7.318

GDP GDP per capita 156 1,539 1,875 6,533 143.8

Bcomp Bank competition 137 0.308 0.130 0.608 -0.592

Bsize Bank size 149 1.867e+06 5.117e+06 2.866e+07 69,115

zscore Z score 149 14.84 8.434 42.71 -4.138

When outliers are excluded such as Nigeria, the Central African Republic, Burkina Faso and Mauritania, who have a much higher bank credit to deposits ratio, the mean decreases to 75%. This underlines the previous findings showing that many banks in SSA are reluctant to provide the private sector with credit. The high standard deviations of both dependent variables reflect the heterogeneity of SSA, which motivates the classification of countries in different income-levels. A look at the explanatory variables shows that the average regulatory quality is relatively low with a negative of -0.4. This highlights the low regulatory structure in SSA. The ownership structure has developed in the last years so that foreign banks became more and more actively involved in banking systems in SSA. The average amount of foreign bank in SSA countries accounts for 63%. However, some financial systems such as the Ethiopian one does not incorporate any foreign banks (0%), whereas the financial system of Zambia consists only of foreign banks (100%).The average amount of government bonds to GDP is 37%. Inflation rates are highly spread; the minimum value is -2% (Senegal) while the maximum value amounts to 44% (Ethiopia). For this reason the inflation rate is transformed to a logarithm. The correlation matrix can be found in the appendix 1, table C.

3.2. Trends in the Data

(17)

overall trend however confirms the findings in the empirical literature. The figure shows a high variation of intermediation efficiency of banks across income levels.

Figure 4: Mean of intermediation efficiency of banks in Sub-Saharan Africa by income levels

Source: Global Financial Development Database (GFDD), World Bank (2013)

Note: Low-income economies consist of 23 countries; lower-middle-income economies of 12 countries, and upper-middle-income economies contain 7 countries.

It can be observed that the intermediation efficiency in upper-middle income economies was low in 1999. On the contrary, lower-middle-income, as well as low-income economies had a higher intermediation level of banks of approximately 78 %. In the beginning of 2001, the trend reversed. By the end of the 1990s banks in upper-middle economies started to provide more and more credit to the private sector. The increasing privatisation of banks by the end of the 90s (for example in South Africa) may have contributed to this upward trend. Over the observed years the level of credit to the private sector has been volatile in all income levels. This evolution is in line with the unstable macroeconomic economic conditions prevailing in many SSA countries.

From the beginning of the 2000s low-income economies have been given out less credit to the private sector compared to other economies. The fact that the growth of bank credit to total deposits stagnated from the 2008 to 2011 for all income levels suggests that the global financial crisis could have had some impact on the financial situation in SSA. This will be elaborated on in section 5. As figure B in the appendix 2 illustrates, lending of banks in SSA to the private sector has been far less compared to other developing economies outside Africa. While in 2011 banks in SSA have been lending 73% of their bank deposits to the private sector, banks in other developing areas such as Latin America and Caribbean (83%), East Asian Pacific (89%) and Europe and Central Asia (118%) provided more credit to the private sector.

40! 45! 50! 55! 60! 65! 70! 75! 80! 85! 1999! 2000! 2001! 2002! 2003! 2004! 2005! 2006! 2007! 2008! 2009! 2010! 2011!

Bank credit to total deposits (%), 1999-2011

(18)

Figure 5: Mean of financial depth in Sub-Saharan Africa by income levels

Source: Global Financial Development Data (GFDD), World Bank (2013)

Figure 5 shows the financial intermediation relative to economic activity, reflected by private credit by deposit money bank to GDP (%) for the years 1999 to 2011. The figure depicts the volume of credit in a financial system, which is a proxy for the financial depth. Although one can observe a slight upward trend in the amount of credit to the private sector for all income levels, the overall ratios remains very low, with an average of 16% and 22% for low-income and lower-middle-income economies. Upper-middle-lower-middle-income economies experienced a small but constant rise of credit to the private sector until 2009, since then the ratio is steady at 40%. The intuition is that the increase in financial depth leads to an increase in the share of deposits invested in productive projects of the private sector fostering economic growth. However, this cannot be observed when looking at both figures. The intermediation efficiency of banks is highly volatile and low for all income levels.

In general credit is pointed out as main ingredient for a functioning banking system. Yet Sub-Saharan Africa has not been able to provide sufficient credit to the private sector, while this is expected to lead to an increase in economic growth (GDP per capita).

3.3. Baseline Model

Following Andrianova’s (2011) and Amidu’s (2014) panel data analysis, the baseline model incorporates a two-step approach in order to estimate determinants of financial depth and the intermediation efficiency of banks in SSA. However, the precondition for an efficient intermediation of banks is a large volume of credit relatively to GDP. Thus, in the first step, the determinants of financial depth (i.e. the volume of credit) are computed as followed:

0! 5! 10! 15! 20! 25! 30! 35! 40! 45! 1999! 2000! 2001! 2002! 2003! 2004! 2005! 2006! 2007! 2008! 2009! 2010! 2011!

Private credit by deposit money bank to GDP (%),

1999-2011

(19)

!"#$%&'_!!"= ! + !!!!"!"+ !!!!"#$%&'_!!"!!+ !!!!_!"#$!"!!+!!!"#!"#!"!!+ !!Χ!"+ !!"! (1)

where !"#$%&!_!!" represents the private credit by deposit money bank to GDP ratio of country c in period t.!!"!" represents the regulatory quality, !"#$%&'_!!"!!reflects the share of foreign bank

assets among total banks asset. !_!"#$!"!! reflects the government bonds issued and !"#$%&!"!!

shows the inflation level.6_{Once the determinants of the volume of credit to GDP are identified, the}

second step elaborates on the factors determining the intermediation efficiency of banks in SSA:

!"#$$#%!"= ! + !!!!"!"+ !!!!"#$%&'_!!"!!+ !!!!_!"#$!"!!+!!!"#!"#!"!!+ !!Χ!"+ !!" (2)

where !"#$$#%!" is the ratio of bank credits to total deposits. Χ!" measures country-specific

characteristics . ! and ! are parameter vectors and !_!" represents the error term. The baseline regression model consists of all selected 28 countries in SSA listed in the appendix.

3.4. Dependent Variables

Private credit by deposit money bank to GDP (%) (Private_C) relates to the general financial

intermediation potential in an economy relative the economic activity (i.e. financial depth). The measure separates credit issued to the private sector from credit granted to government or the public sector (Levine, 2004). The former recipients of credit are assumed to utilize credit in a more productive way than the government or the public sector (King & Levine, 1993). The data is extracted from FDSD.

Bank credit to total bank deposits (%) (Creddep) explicitly gives information about the

intermediation efficiency of banks. As for the private credit to GDP ratio, bank credit to total bank deposits only lists credit provided to the private sector and not to the government or the public. By definition, the higher the share of deposits used for credits to the private sector, the lower the level of liquid assets. A lower level of liquid assets is attributed to a higher efficiency of banks.

3.5. Explanatory Variables

Overall, major determinants of financial depth and intermediation efficiency are external to the bank. This is especially the case for lower-income countries such as the majority of SSA. As many studies emphasize, the environment in those countries is a major factor hindering banks from lending a big portion of their deposits to the private sector (Andrianova, Baltagi, Demetriades, & Fielding, 2014), (Demirgürc-Kunt, 2006), (Demetriades & David, 2009). The explanatory variables are chosen based on that fact. As discussed before, the legal structure, ownership structure and

(20)

macroeconomic conditions are seen as major determinants of the intermediation efficiency of banks.

Regulatory quality (RQ) serves as a proxy for the legal structure. Regulatory quality is among the

measures of the Worldwide Governance Indicators (WGI) database, which sum up the perception of economic agents on the quality of governance in emerging and developing countries. The regulatory quality represents the perception of economic agents of “the ability of government to formulate and implement sound policies and regulations that permit and promote private sector development” (Kaufmann, Kraay, & Mastruzzi, 2010). A functioning legal structure is necessary to avoid information asymmetries, which result in moral hazards and adverse selection. These problems are common in SSA where regulatory measures are poorly defined and enforcement of the law is weak. The estimates of regulatory quality range between -2.5 (weak) and 2.5 (strong) governance performance.

Hypothesis 1: Regulatory quality is expected to have a positive correlation with financial depth and the intermediation efficiency of banks.

The foreign bank assets among total bank assets (%) (Foreign_A) is a proxy for the ownership

structure of banks. The impact of foreign banks on the level of credit to the private sector can be derived either by calculating the ratio of foreign bank assets to total assets or by taking the deposit share of foreign bank in an economy (Rashid, 2011), (Claessens & van Horen, 2012). Due to a high amount of unreported values of bank total deposits the ratio of foreign bank deposits cannot be derived. Therefore, foreign bank assets to total assets is used instead. The goal is to show how foreign banks penetration into the domestic banking system affects financial depth and the intermediation efficiency of banks. Especially for the years before 2003 there is a lack of ownership information of banks, which inhibit to track the shareholders holding more than 50% of a bank. Thus, this study refers to the ratio of foreign bank assets among total assets to proxy for foreign bank presence. Foreign banks are expected to have a comparative advantage in the service sector. Furthermore, it is assumed that foreign banks prefer to invest in liquid assets due to information asymmetries in SSA financial systems. Based on that, they will lend only a limited amount to the private sector. Since most foreign banks outcompete domestic banks, domestic banks will reduce the amount of credit to the private sector as well. The data on foreign bank assets among total bank assets provided by Claessens and van Horen (2012) restricts the model to the years 2004 to 2011.

(21)

Moreover, there is evidence that macroeconomic conditions influence the financial performance of banks and therefore the intermediation efficiency.

Government bonds (%) (G_debt) is measured by the amount of government debt issued in the

domestic economy as a percentage of GDP. The corresponding data is extracted from The World Economic Outlook (WEO) database of the International Monetary Fund. There is empirical evidence that investment in government bonds crowd-out credit to the private sector (Freedman & Click, 2006) (Christensen, 2004). In many countries, banks prefer to invest in government bonds since they are assumed to be more secure than granting credit to the private sector. Freedman and Click (2006) conclude that “The availability of government debt that offers moderate or high returns renders bank less inclined to search for profitable lending opportunities with private sector borrowers”(Freedman and Click, p.292).

Hypothesis 3: The higher the amount of government bonds issued in the domestic economy, the lower the level of credit to the private sector, i.e. the lower the intermediation efficiency.

Inflation (%) (loginf) serves as a proxy for financial stability and is reflected through the measure

of annual percentage of average consumer prices on year-on-year changes. In general, banks want to maintain low inflation levels in order to maintain price stability. However, many countries in SSA experience high inflation rates. There is empirical evidence that financial instability, which is consistent with high inflation rates, is negatively and significantly correlated with financial development (Andrianaivo & Yartey, 2009) (Boyd, Levine, & Smith, 2001). According to previous findings, high inflation increases adverse selection problem in the credit market. As a consequence, financial institutions provide fewer loans. Inflation can lead to an inefficient allocation of credit, which in turn can lead to a higher probability of loan defaults. The data on inflation is obtained from the World Economic Outlook database (2015) of the International Monetary Fund.

Hypothesis 4: Higher inflation reduces intermediation efficiency.

Controls

This research utilizes several control variables to capture country-specific characteristics (!!Χ!").

(22)

an incentive to increase their investment in credit the private sector. The data on GDP per capita growth is extracted from the Global Financial Development Database (GFDD) of the World Bank (2013). Furthermore, banking structure is proxied through Bank size (Bsize) reflecting the average of total assets of banks in a country and bank competition (Bcomp), which is measured by the Lerner Index. The probability that a country’s commercial banking system defaults is captured by the Z-score (zscore). The FDSD, GFDD and Bank Scope provide data for all control variables.

3.6. Tests

Figure 6 shows the plots of the key dependent variables. Plots are useful to check potential influential observations and to detect outliers in the dataset causing biased results. The x-axis displays the leverage, which is plotted against the normalized residual squared. Leverage gives information of how far an observation is from the mean. An observation is considered as influential if excluding it from the regression causes a significant change in the regression coefficient. This is measured by the Cook’s distance (Cook’s D), which pulls together information about the leverage and residuals. Outliers with a Cook’s D above 1 have a large influence on the regression results.

Figure 6: Cook’s D of key dependent variables

Figure 6 shows that the outliers differ with respect to the dependent variable. Clear outliers in the plot with the dependent variable Creddep are particular observations of Nigeria, Burkina Faso, Malawi and South Africa. However, only Kenya, Madagascar, and Malawi exhibit a Cook’s D above 1, which implies that they have a strong influence on the results. Therefore these observations are excluded from the baseline regression model. The plot with the dependent variable Private_C identifies Liberia, Mauritius, and South Africa as outliers. Since all countries have a Cook’s D close to zero they are not influential and therefore kept in the baseline model. Due to missing values in the sample omitting several countries would substantially restrict the number of observations and one would run the risk of having inconsistent results.

Angola Angola Angola Angola Angola Angola AngolaAngolaAngola_Angola_Angola_Angola

Benin Benin Benin Benin Benin Benin Benin Benin

BeninBeninBeninBeninBotswanaBotswanaBotswanaBotswanaBotswanaBotswanaBotswanaBotswanaBotswanaBotswanaBotswanaBotswanaBotswana Burkina FasoBurkina FasoBurundiBurkina FasoBurkina Faso

Cameroon Cameroon Cameroon Cameroon Cameroon Cameroon Cameroon Cameroon Cameroon

CameroonCape VerdeCape VerdeCape VerdeCape Verde Central African Republic

ChadChadChad Chad Chad Chad Chad Chad Congo, Dem. Rep. Congo, Dem. Rep. Congo, Dem. Rep. Congo, Dem. Rep. Congo, Dem. Rep.Congo, Dem. Rep. Congo, Rep.

Cote d'IvoireCote d'IvoireCote d'IvoireCote d'IvoireCote d'IvoireCote d'IvoireCote d'IvoireCote d'IvoireCote d'IvoireCote d'IvoireCote d'IvoireCote d'Ivoire EthiopiaEthiopiaEthiopiaEthiopia Ethiopia EthiopiaEthiopia Ethiopia Ethiopia EthiopiaGabonGabonGabonGabon

Gabon Gabon

Ghana Ghana_Ghana Ghana Ghana GhanaGhanaGhana GhanaGhanaGhana Guinea Guinea GuineaGuineaGuinea GuineaGuinea Guinea-BissauGuinea-BissauLesothoLesothoKenyaKenyaKenyaKenyaKenyaKenyaKenyaKenyaLesothoKenyaKenyaKenyaKenyaKenya Liberia

Liberia

Liberia Liberia

Madagascar_MadagascarMadagascar Madagascar MadagascarMadagascar Madagascar Madagascar Madagascar MadagascarMali_Mali_MaliMaliMaliMalawiMaliMaliMaliMaliMaliMalawiMalawiMalawiMalawiMalawiMalawiMalawiMalawi Mali

Mali

MauritiusMauritiusMauritiusMauritius MauritiusMauritiusMauritiusMauritius Mauritius MauritiusMauritius Mauritius Mozambique

MozambiqueMozambiqueMozambiqueMozambique_MozambiqueMozambiqueMozambiqueMozambiqueMozambique MozambiqueMozambiqueMozambiqueNigerNigerNigerNigerNigerNigerNigerNigerNigerNigerNigerNamibiaNigerNigerNamibiaNamibiaNamibiaNamibiaNamibiaNamibiaNamibia NigeriaRwanda_RwandaNigeria NigeriaNigeria NigeriaNigeria Rwanda Senegal Senegal Senegal Senegal Senegal Senegal SenegalSenegal Senegal SenegalSenegalSenegal

Sierra Leone

Sierra Leone Sierra Leone

Sierra Leone_{Sierra Leone} _{South Africa}_{South Africa}_{South Africa}_{South Africa}South Africa_{South Africa}_{South Africa}_{South Africa}_{South Africa} Swaziland

SwazilandSwazilandSwaziland Swaziland SwazilandSwaziland SwazilandSwazilandSwazilandTanzaniaSwazilandSwazilandTanzaniaTanzaniaSwazilandTanzaniaTanzaniaTanzaniaTanzaniaTanzania Togo TogoTogo Togo Togo Togo Togo TogoTogoTogoTogoZambiaZambiaUgandaUgandaUgandaUgandaUgandaUgandaUgandaUgandaZambiaUgandaUgandaZambiaUgandaZambiaZambiaUgandaZambiaUganda

0 .1 .2 .3 .4 L e ve ra g e 0 .01 .02 .03 .04 Normalized residual squared

Plot with dependent variable: Private_C

Angola Angola AngolaAngolaAngola AngolaAngola

Benin Benin Benin

Benin

BeninBotswana BotswanaBotswana BotswanaBotswanaBotswana

Burkina Faso Burkina Faso Burkina Faso Burkina Faso Burundi Burundi Burundi Burundi _Burundi Burundi Burundi Cameroon Cameroon CameroonCameroon CameroonCameroonCameroon

Cote d'IvoireCote d'Ivoire_{Cote d'Ivoire} Ethiopia Ethiopia

Ethiopia Ethiopia

Ghana Ghana_{Ghana Ghana}

Kenya KenyaKenya KenyaKenya KenyaKenya Madagascar Madagascar Madagascar MadagascarMadagascar MadagascarMadagascar Malawi Malawi Malawi Malawi Malawi

MalawiMaliMali _Mali Mali

Mali Mali Mauritius Mauritius_MauritiusMauritius

Mauritius Mauritius Mauritius Mozambique Mozambique Mozambique MozambiqueMozambique Niger Niger Niger Niger NigeriaNigeria Nigeria NigeriaNigeria Rwanda Senegal Senegal

SenegalSenegalSenegal _Senegal South Africa South Africa South Africa South Africa South AfricaSouth Africa

Swaziland SwazilandSwaziland Swaziland Swaziland

Swaziland

Tanzania Tanzania Tanzania_Tanzania_Tanzania_Tanzania

Togo Togo Togo Togo Togo Togo Uganda UgandaUgandaUganda_{Uganda Zambia}_Zambia

Zambia Zambia ZambiaZambiaZambia

0 .1 .2 .3 .4 L e ve ra g e 0 .02 .04 .06 .08 Normalized residual squared

(23)

This study utilizes a panel data approach for the model specification. The advantage of a panel data analysis is that additionally to the cross-sectional analysis it also exploits the time-series variation in the dataset. Moreover, a panel data estimation method prevents biases associated with cross-country regressions such as potential unobserved cross-country-specific effect. As part of the error term, these effects can lead to a correlation between the error term and explanatory variables. This in turn generates inconsistent results.

The panel dataset used in this study is unbalanced because of missing observations for certain countries and particular years, which is a drawback when analysing SSA. A panel data analysis can be performed in different ways. Generally, one has to decide whether a pooled OLS, a random or fixed effects estimation is appropriate for the analysis. In a pooled OLS model the intercept is expected to be the same for all individuals and all time periods. The assumption is that the errors for different individuals and time are uncorrelated. However, as already discussed a large heterogeneity across Sub-Saharan African countries exists, implying that the OLS estimation is inconsistent. In this regard the Breusch-Pagan Lagrangian multiplier test helps to identify whether random effects are present. The test is computed for every regression and reveals a p-value smaller than 0.05 stating that random effects are present.

In this case a random effects or fixed effects model is appropriate for the panel data analysis. The fixed effects model assumes that all differences between individuals are captured by the intercept. The intercept (i.e. fixed effects) captures “for individual-specific, time-invariant characteristics” (Hill, Griffiths, & Lim, 2011). However, the random effects model treats the heterogeneity across individuals as a random component. To decide whether the fixed effects or the random effects model is appropriate for the analysis, the Hausman test is conducted. It compares the estimates from both models. Random effects and fixed effects model are both appropriate if there is no correlation between the error component ui and independent variables. If it turns out that ui is

correlated with the explanatory variables then there is evidence in favour of the fixed effects model over the random effects model. The null hypothesis is that ui is uncorrelated with any explanatory

(24)

over the random effects specification. Hence, the fixed effects model is applied to all regressions. The random effects estimation results are reported in the appendix.

Furthermore, a time-dummy test is conducted to look for temporal variation in the dependent variables that is not already captured by the explanatory variables. The equation with private credit to GDP shows a significant p-value, which indicates that the explanatory variables do not fully capture temporal variations in the dependent variable. With regard to the regression including bank credit to total deposits, time dummies are not needed. Though, time-dummies are included for both regressions to account for time-fixed effects.

Figure 6 shows evidence of heteroskedasticity for both estimations, which suggests that the error terms are correlated over time within a country. The likelihood-ratio test confirms the presence of heteroskedasticity in both equations. Since heteroskedasticity leads to biased standard errors robust standard errors are applied in all regressions.

Next, a test is conducted to examine whether there are patterns in the residuals, which would imply that they are autocorrelated. However, if the presence of autocorrelation is ignored, the coefficients are inefficient. This means that the error term has a lot of variation, increasing the likelihood of drawing false conclusions. The Wooldridge test for autocorrelation shows evidence for autocorrelation in all regressions. Thus, robust standard errors are included in both equations to control for autocorrelation. Moreover, the lagged values and the time dummies in the model additionally help to account for autocorrelation.

The presence of multicollinearity in the regressions might lead to an upward bias of the regression results. Therefore a test for multicollinearity is conducted. The test calculates the uncentered variance inflation factors (VIFs) to detect collinearity of the regressors with the constant term. If VIF is 10 or higher multicollinearity is present and robust standard errors have to be applied to account for it. None of the equations in the baseline model show evidence of multicollinearity. To identify the effects of each control variables on the overall regression results the control variables are incrementally included in the regression.

3.7. Endogeneity

(25)

logged inflation rate as well as government bonds. With respect to private credit to GDP only regulatory quality is subject to endogeneity. The one-period lags already applied in the baseline model help to deal with endogeneity since it is unlikely that the present level of financial depth and intermediation efficiency of banks affects the past values of the explanatory variables.

Nonetheless, to ensure robustness an instrumental variable approach is conducted. The instrumental variable (IV) two stage least squares (2SLS) fixed effects estimator is applied for both regressions (see equation 3 and 4). However, the commonly used the Arellano-Bond dynamic-panel GMM estimator is not suitable for this dynamic-panel data. For the GMM estimator to be consistent the number of individuals has to be very large (N), which does not apply to this panel data. Thus, “the cluster-robust standard errors and the Arellano-Bond autocorrelation test may be unreliable” (Roodman, 2009). Hence, the instrumental variable 2SLS approach is utilized. For instrumental variables to be reasonable they have to fulfil two requirements. First, they should not be correlated with the regression error term and second, the IV should be strongly correlated with x (endogenous explanatory variable) (Hill, Griffiths, & Lim, 2011). Based on that, lagged values of the endogenous variables and first differences are considered as instrumental variables. Since the logged inflation rate and government bonds are already lagged in the baseline model they are instrumented using a second-period lag. Another reason to apply lagged values is that it is unlikely to find instruments, which influence macroeconomic factors but at the same time do not influence the financial situation of a country. Moreover, data is strongly restricted. This approach is common in the existing literature examining determinants of financial development and intermediation efficiency of banks (Iossifov & Khamis, 2009).7

In order to apply the selected instruments at first, the strength of the IV is tested by a F-joint test of significance. After the null hypothesis stating that the instrumental variables are ‘weak’ is rejected for both regressions, the validity of those instruments needs to be tested. Therefore the Sargan test (overidentification test of all instruments) is computed (Hill, Griffiths, & Lim, 2011). With a p-value of 0.05 the null hypothesis is not rejected and the instruments are considered as valid for both regressions. Thus, the lagged values and lagged differences of the selected explanatory variables are valuable instruments to account for endogeneity. According to the endogeneity test, GDP_growth is not endogenous. However, several studies have shown that the causality between GDP growth and financial development can run in both ways (Yay & Oktayer, 2009) (Fernandez de Guevara & Maudos, 2007). While losing significance when included as endogenous variables in the IV 2SLS estimation, GDP_growth does not change the overall results.

7

(26)

4. Regression Results

4.1. Regression Results of the Baseline Model and IV 2SLS estimator

Table 2 and 3 represent the results of the baseline model under the fixed effects estimation (column 1-6). Column (7) and (8) depicts the outcome of the robustness check (IV 2SLS). The explanatory variables Foreign_A, G_debt, and loginf are transformed in one period lags. The intuitive assumption is that effects on financial indicators such as on the key dependent variable are visible in a subsequent time period. Once the control variable on bank competition is entered, there is a loss of sample observations in the regression.

Table 2 shows the results for the determinants of private credit to GDP. The results suggest that out of the four main explanatory variables, regulatory quality and foreign bank presence have a significant impact on financial depth. Regulatory quality is positively and significantly correlated with the dependent variable, at a 1% significance level. This is consistent with the assumption that better legal structure increases the volume of credit to GDP in SSA financial systems.

Table 2: Fixed effects Model and IV 2SLS estimation of Determinants of Private credit to GDP

(1) (2) (3) (4) (5) (6) (7) (8)

Private_C Fixed effects model IV 2SLS

RQ 7.236 8.086 11.48** 11.48** 11.43** 6.455** 24.02*** 17.64** (4.645) (4.833) (4.424) (4.544) (4.413) (2.998) (7.727) (7.552) L.Foreign_A -0.0765*** -0.0770*** -0.0864** -0.0864** -0.0887** -0.0800** -0.116*** -0.102** (0.0262) (0.0270) (0.0321) (0.0325) (0.0341) (0.0331) (0.0411) (0.0392) L.G_debt 0.0187 0.0130 0.0152 0.0152 0.0130 0.00662 0.0108 0.0107 (0.0118) (0.0123) (0.0152) (0.0154) (0.0150) (0.0109) (0.0182) (0.0140) L.loginf -0.648* -0.576* -0.498 -0.498 -0.418 -0.473 -0.325 -0.449 (0.333) (0.322) (0.347) (0.367) (0.385) (0.295) (0.434) (0.337) GDP_growth -0.245** -0.306*** -0.306*** -0.324*** -0.369*** (0.0978) (0.0898) (0.0930) (0.0909) (0.0676) Bcomp -2.879 -2.876 -1.379 -4.360 -5.260** -7.125*** (4.936) (5.312) (5.060) (3.035) (2.540) (2.283)

Bsize -1.88e-09 -6.76e-08 -1.96e-07 3.34e-07 2.43e-07

(3.43e-07) (3.16e-07) (4.26e-07) (3.69e-07) (3.87e-07)

zscore -0.156 0.00654 -0.148 -0.00340 (0.146) (0.135) (0.154) (0.143) GDP 0.00729*** 0.00528** (0.00250) (0.00225) Constant 25.98*** 27.39*** 30.25*** 30.25*** 32.58*** 18.73*** (2.325) (2.347) (3.632) (3.614) (5.253) (6.186) Observations 154 154 135 135 135 135 134 134 R-squared 0.503 0.534 0.559 0.559 0.564 0.587 0.489 0.532 Number of country 27 27 25 25 25 25 24 24

Time-fixed effects Yes Yes Yes Yes Yes Yes Yes Yes

Robust standard errors in parentheses

(27)

The lagged value of foreign bank presence is negatively correlated with financial depth, at 1 % significance level. This means that a one percentage point increase in foreign bank assets as share of total bank assets reduces the volume of credit to GDP by 0.08 percentage points. This result is consistent with theory. The issuances of government bonds as well as the inflation level do not have a significant impact on financial depth in SSA. The inclusion of several control variables does not change the overall results. Since the variable GDP_growth shows a significant and negative effect on both dependent variables, including GDP per capita as proxy for economic development tests if the relationship between economic development and the dependent variables holds (Column 6 and 8). Exchanging GDP per capita growth with GDP per capita does not change the overall result and significance of the model. However, GDP per capita is positively and significantly correlated with the dependent variables, whereas GDP per capita growth has the opposite effect on the dependent variables. This outcome underlines the mixed results in the previous literature with respect to the relationship between economic development and financial development.

As reported in section 3.7 the transformation of the explanatory variables suspected to be endogenous in one-period lag values accounts for endogeneity. Nonetheless, in order to rule out potential biases in the regression results the IV 2SLS is applied (column 7 and 8). The results show no significant difference to the fixed effects estimation. The coefficients of regulatory quality and foreign bank presence are considerably larger in the IV 2SLS estimation. Thereafter, a one unit increase in regulatory quality leads to a 24-percentage points increase in the volume of credit (from initial 11 percentage points). The coefficient suggests an economically strong influence of regulatory quality on private credit to GDP. Moreover, a one percentage point increase in foreign bank presence leads to a reduction of private credit to GDP by 0.1 percentage point (from initial -0.09 percentage points). The increased coefficients suggest that the IV 2SLS regression removes the variation in the explanatory variable caused by the error term. Overall, the results are plausible and suggest that less foreign bank activities and a better legal structure have the expected influence on private credit to GDP.

What determines Financial Depth and Intermediation Efficiency of banks in Sub-Saharan Africa?