Faculty of Economics and Business
Business Economics: Managerial Economics and Strategy
MASTER THESIS
The Effect of Islamic Banking
Development on the Banking Sector and
Economic Growth
Author: Krist´yna Drmotov´a Student Number: 11373687 Supervisor: dr. Adam S. Booij Academic Year: 2016/2017
Declaration of Authorship
The author hereby declares that he compiled this thesis independently, using only the listed resources and literature.
The author grants to University of Amsterdam permission to reproduce and to distribute copies of this thesis document in whole or in part.
Amsterdam, July 7, 2017
Acknowledgments
I am especially grateful to my supervisor, dr. Adam S. Booij, for valuable advice and useful suggestions. I am also grateful to my family and friends for an infinite support.
Abstract
This thesis investigates the impact of the development of Islamic banking on the overall banking sector development as well as on the GDP growth. A differencin-differences model is designed to assess the hypotheses, being es-timated by the fixed-effects estimation technique with heteroskedasticity- and autocorrelation-robust clustered standard errors. Using a sample of 49 coun-tries during the period 2007–2014, from which 22 councoun-tries have Islamic bank-ing established, this thesis reports a negative relationship between the penetra-tion level of Islamic banks in the banking sector and the development of this banking sector. Further, it finds no significant effect on the GDP growth.
JEL Classification G21, O16
Keywords Islamic finance, Islamic banking, financial
de-velopment, banking system structure, economic growth
Author’s e-mail kristynadrmotova@seznam.cz
Contents
List of Tables vi
List of Figures vii
1 Introduction 1
2 Review on Islamic banking 3
2.1 The basics of Shari’ah . . . 3
2.2 Basic principles of an Islamic financial system . . . 4
2.3 History and magnitude of Islamic finance . . . 5
2.4 Islamic banking and finance products . . . 7
2.5 Pros and cons of Islamic banking compared to conventional bank-ing . . . 7
3 Recent empirical studies 9 4 Empirical part 12 4.1 Methodology and data description . . . 12
4.1.1 Methodology . . . 14 4.1.2 Description of variables . . . 18 4.2 Empirical results . . . 22 4.2.1 Hypothesis 1 . . . 22 4.2.2 Hypothesis 2 . . . 26 4.3 Robustness checks . . . 29 4.4 Discussion of results . . . 32 5 Conclusion 35 Bibliography 40 A Appendix I
List of Tables
4.1 Summary statistics . . . 23 4.2 Regression tests of the first hypothesis . . . 25 4.3 Regression tests of the second hypothesis . . . 27 4.4 Second robustness check for banking sector development: 2011–
2014 . . . 29 4.5 Second robustness check for the GDP growth: 2011–2014 . . . . 31 A.1 First robustness check for the banking sector development:
List of Figures
4.1 Histogram of the penetration level of Islamic banking in the over-all banking sector . . . 19 4.2 Cumulative distribution function of the penetration level of
Chapter 1
Introduction
Finance that is in line with Islamic beliefs needs to be subject to the core tenet of Islamic religion, the Shari’ah law. As Islam prohibits any form of interest, the so called riba, the classical interest-based financial system fails to fulfil the needs of faithful Muslims. Thus, Islamic finance relies on an alternative, the profit-and-loss sharing, which makes it different to the conventional banking in many aspects. Even though the core idea of Islamic finance is as old as the religion itself, the Islamic banks have been established only recently. The first one was opened in Mit-Ghamr, Egypt, in 1963. Since then Islamic banking has found its place in numerous countries. In some of them, namely Iran and Sudan, the banking system even operates fully on the Shari’ah-compliant basis. After the recent financial crisis the Islamic banking assets have grown at double-digit rates and are expected to continue so in the following years (EY 2016). Given that 24.1 % of the worldwide population is Muslim (Lipka & Hackett 2017), the potential of Islamic finance is not negligible.
Many researchers have centered their attention to the comparison of Islamic to conventional banking but its impact on the macroeconomic indicators has not yet been exhaustively investigated. The outcome of these studies is, how-ever, very relevant for the policy-makers since they have power in their hands to alter the legal setup with supervisory and regulatory reforms and thereby either facilitate or hinder the further development of Islamic banking. Be-cause of its features it might deepen the financial inclusion via attracting the previously “unbanked” individuals. Moreover, its development may spur com-petition and the need of innovation within the banking sector as new products become available on the market. On the other hand, due to the lack of liquid and hedging instruments, it is possible that Islamic banking cannot compete
1. Introduction 2
with the conventional banking (Beck et al. 2013; Imam & Kpodar 2016) Therefore I attempt to contribute to the existing literature by examining the effect of Islamic banking development on the banking sector development with the differences-in-differences model. Furthermore, already since 1911 Schum-peter & Backhaus (2003) argued that financial development has a crucial role in economic development because the financial sector decides on the reallo-cation of resources. This positive relationship between the two variables was supported by numerous authors, e.g. Beck et al. (2000). Thus, I also study the effect of Islamic banking development on the GDP growth, using the same model.
The analysis is performed on the panel data sample comprising of 49 coun-tries with at least 20% Muslim population (except for Thailand which, however, has a Muslim population of almost 4 millions) over the span of 8 years, specif-ically from 2007 to 2014. Out of that, Islamic banking is established in 22 countries. The fixed-effects estimation technique with heteroskedasticity- and autocorrelation-robust clustered standard erros is used to assess the hypotheses of the effect of Islamic banking development.
The results suggest a negative effect of the Islamic banking development on the overall banking sector development, as opposed to my hypothesis. More specifically, the higher the level of Islamic banking development, the larger is the expected difference between the banking sector development of that country and of the country where Islamic banking is not established, given there is no other difference between the two countries than the presence of Islamic banking. The second hypothesis of positive effect of Islamic banking develoment on the GDP growth was also not supported by my analysis. The estimated effect of the Islamic banking development on the GDP growth is not statistically different from zero.
Structure of the thesis is as follows. Chapter 2 is devoted to the overview of the Islamic banking. Firstly the basics of Shari’ah and Islamic financial system are explained. Then the history and magnitude of Islamic banking is discussed followed by the outline of the Islamic banking and finance products. The chapter is closed with the elaboration on the pros and cons of Islamic banking compared to the conventional counterpart. Chapter 3 is devoted to the review in the recent empirical literature. The empirical part of this thesis is discussed in chapter 4 starting with the methodology and data description. Then the results are presented, followed by the robustness checks. The last section is devoted to discussion of the results. Chapter 5 concludes the thesis.
Chapter 2
Review on Islamic banking
In this chapter, the basic principles of Shari’ah law and the Islamic banking will be explained. To understand the current state and future development of the Islamic banking, it is important to know the past. Thus, a part of this chapter is devoted to the history and magnitude of Islamic finance. Then I provide an outline of Islamic banking and finance products and close the chapter with the assessment of pros and cons of Islamic banking compared to the conventional counterpart.
2.1
The basics of Shari’ah
Muslim’s life is commanded by three different sets of rules. Adiqah, or faith, represents the core relationship between people and the Creator. Shari’ah, or law, advises on how to transform and demonstrate the faith and beliefs into action and daily practices. The third set, akhlaq, constitutes Muslim’s behaviour, attitude, and work ethics in society (Iqbal & Mirakhor 2011).
Shari’ah then comprises two components: iquadat and muamalat. Iquadat are the rituals that help an individual to reach his inner understanding of his relationship with Allah. Muamalat defines the rules for commercial, financial, and banking systems and thereby determines the conduct of economic activities (Iqbal & Mirakhor 2011).
As follows, finance that is in line with Islamic beliefs needs to be subject to the core tenet of Islamic religion, the Shari’ah law. It is based on Divine guidance preached by the two most-important sources of the Islamic faith, the Qur’an and the sunnah, and includes five basic objectives: to protect the faith, life, progeny, property, and reason (Iqbal & Mirakhor 2011). The market
con-2. Review on Islamic banking 4
duct is built on five pillars comprising “property rights, free flow of information, trust, contract, and the right not to be harmed by others, and the obligation not to harm anyone” (Iqbal & Mirakhor 2011, pg. 39).
Some of the main features of Islamic finance are the prohibition of riba, or interest, of gharar, i.e. ambiguous contracts or deals without full information disclosure, and of other forms of exploitation like qimar, or gambling, and mysir, i.e. chance-based games that involve deception (Iqbal & Mirakhor 2011).
2.2
Basic principles of an Islamic financial system
“Those who consume interest cannot stand on the Day of Resurrection except as one stands who is being beaten by Satan into insanity. That is because they say: ‘Trade is just like interest.’ But Allah has permitted trade and has forbidden interest. So whoever has received an admonition from his Lord and desists may have what is past, and his affair rests with Allah. But whoever returns to dealing in interest or usury — those are the companions of the Fire; they will abide eternally therein.”
Quran (2:275) Islam encourages profit earning as it ex post follows creation of wealth and successful business. On the contrary, interest that is specified ex ante is a cost exerted independent of the outcome of the entrepreneurial activity and it may not create additional wealth (Iqbal & Mirakhor 2011).
However, Qur’an does not distinctly define what riba is as it was not of a big concern at the time. There is a common consensus among scholars, though, that riba is “the practice of charging financial interest or a premium in excess of the principal amount of a loan” (Iqbal & Mirakhor 2011, pg. 57). The ban of interest thus does not encompass only usury but interest on debt in any form (Mohieldin 2012). The forbidden interest is specified by at least four characteristics: “(1) it is positive and fixed ex ante; (2) it is tied to the time period and the amount of the loan; (3) its payment is guaranteed regardless of the outcome or the purposes for which the principal was borrowed; and (4) the state apparatus sanctions and enforces its collection” (Iqbal & Mirakhor 2011, pg. 59).
Moreover, the Islamic fundamental sources do not state any complete rea-soning behind this prohibition. Rationale that scholars voice most often is that riba is a form of social and economic exploitation which is in conflict with
2. Review on Islamic banking 5
the Islamic objectives of social justice. Furthermore, interest on money loaned represents property rights claim that is unjustifiable and instantaneous in the Islamic interpretation (Iqbal & Mirakhor 2011).
Therefore, Islamic finance relies on an alternative to the financial system based on interest-rate debt, i.e. on risk-and-reward sharing. Since Islam puts emphasis on social justice, equality, and property rights, it requires that bor-rowers and lenders share rewards as well as losses in an impartial fashion. Supplier of funds thus becomes investor rather than creditor. The risk-sharing principle builds on a saying of the Prophet that “profit comes with liability”, in other words one is entitled to profit only if he faces the liability, or risk of loss (Iqbal & Mirakhor 2011).
Since borrowers and investors share the outcome no matter whether it is profit or loss, they are discouraged to take on excessive risk and care more about long-term goals. All financial contracts should be backed by assets or transactions in the real economic sector. As follows, the two sectors are closely linked and the real values of assets and liabilities are equal at any time. Fi-nancial institution as an investor commits itself to this business partnership with the borrower and thus it is in its interest to help the borrower find his way out of the bad times. This prevents him from being forced to sell assets at “fire-sale” prices and cascading defaults become more unlikely (Mohieldin 2012).
2.3
History and magnitude of Islamic finance
The inception of the Shari’ah-compliant-services initiatives dates back to the Suez Canal construction even though Islamic finance is as old as the religion itself. Barclays Bank opened its Cairo branch in the 1890s to process the related financial transactions. The criticism of its interest related practices spread fast across the Muslim population in Arab regions and Indian sub-continent. That lifted the wave of scholars offering theoretical alternatives to interest-based banking followed by the opening of the first modern Islamic bank in 1963 in Mit-Ghamr, Egypt (Iqbal & Mirakhor 2011). It managed to mobilize funds through savings accounts where no interest was paid to the holders but instead they were offered small short-term interest-free loans for productive purposes. These funds were invested on the profit-and-loss-sharing basis with entrepreneurs (Abedifar et al. 2015). First commercial Islamic bank, Dubai Islamic Bank, was then opened in 1974 (Iqbal & Mirakhor 2011).
2. Review on Islamic banking 6
One of the key points in the history of Islamic banking development is the foundation of the international financial institution, Islamic Development Bank, in 1975 by the representatives of the Organization of the Islamic Conference member countries. Its intention is to promote economic and social development in the Muslim countries and has functioned as its main financier and promoter (Abedifar et al. 2015).
Since then Islamic banking spread across many countries having its place in numerous states. At this date, Islamic finance assets are densely concentrated in the Middle East and Asia. Iran and Sudan even operate fully Shari’ah-compliant banking systems since 1983 and 1984, respectively (Abedifar et al. 2015). The nine core markets are Bahrain, Qatar, Indonesia, Saudi Arabia, Malaysia, United Arab Emirates, Turkey, Kuwait, and Pakistan and together they account for 93 % of the industry assets. Their Islamic banking assets were estimated to exceed US$ 920 billion in 2015 (EY 2016).
The perspectives for further growth of Islamic finance are also generally positive in spite of the macroeconomic and political environment in many of the emerging markets. The demand is still significantly higher than supply (EY 2016). In most Muslim countries the access to financial institutions and products is relatively low and as the Islamic banks are expected to act as complements to the conventional banking, they may convince many lenders and borrowers to use formal instead of informal markets (Abedifar et al. 2016). Moreover, 10 out of the 25 rapid-growth markets have large Muslim population and these are expected to be an engine of the global growth over the next decade (EY 2016). The intense retail demand and proactive government support help the Islamic banks to outpace in growth their conventional counterparts. The sector has potential for future growth, in particular in countries at a low level of penetration of Islamic banking assets. As an example, “over the last three years, Oman’s Islamic banking sector has gone from zero to an aggregate of around 10% of banking system financing assets as of June 2016” (Ambrose 2016).
Even though Islamic financial assets comprise only under 1% of total global financial assets, they have grown faster than conventional banking since the financial crisis of 2007–2008 (Abedifar et al. 2015). Islamic financial assets had been estimated to have grown at 17.3% annual growth rate between 2009 and 2014 (Islamic Financial Services Board 2015). EY (2016) expects a compound annual growth rate of 14% through 2015-2020 and that the total assets will reach US$1.8 trillion across the nine core markets by the end of 2020.
2. Review on Islamic banking 7
2.4
Islamic banking and finance products
Most of the Shari’ah-compliant products are based on the profit-and-loss shar-ing. Mudaraba contracts specify that profits are shared at a predetermined ra-tio, whereas the bank bears all the losses. The entrepreneur is thus effectively covered by limited liability provisions. He has to ask for approval regarding the major investment decisions. On the other hand, under Musharaka contracts, bank acts as one of several investors and the profits and losses are shared among all investors. Depositors thus become residual claimants instead of creditors (Beck et al. 2013).
The Shari’ah-compliant finance, nonetheless, also offers products not based on profit-and-loss sharing. One example is the Mudaraba contract similar to leasing contract in conventional banking. Since the bank purchases the goods, it evades the prohibition of interest on money lending (Beck et al. 2013). The Islamic financial markets contain not only Islamic banks, but also Islamic win-dows, Islamic insurance (takaful ), capital markets, and other non-bank financial institutions (Iqbal & Mirakhor 2011). Islamic banking is the largest compo-nent of the Islamic market (Islamic Financial Services Board 2015). Sukuk, i.e. the issuance of certificates of trust commonly referred to as Islamic bonds, constitute assets worth US$ 294 billion (Ambrose 2017).
2.5
Pros and cons of Islamic banking compared to
conventional banking
Numerous studies focus on the performance comparison of Islamic to conven-tional banks. The theoretical predictions are ambiguous in many aspects. Firstly, as the Shari’ah-compliant deposits are similar in principle to equity, the depositors’ incentives to monitor and discipline the bank may increase but at the same time banks might be less motivated to monitor and discipline bor-rowers since there is no risk of immediate withdrawal. Also, monitoring and screening costs may be lower in the case of Islamic banks due to smaller agency problems, but on the other hand the higher complexity of Shari’ah-compliant contracts may result in lower efficiency due to higher costs (Beck et al. 2013). Moreover, Islamic banking is a nascent industry and the cost structures might not yet be minimized as they cannot completely exploit economies of scale (Imam & Kpodar 2016).
2. Review on Islamic banking 8
Further, as mentioned above, each transaction has to be performed against a transaction in real economic sector involving tangible assets. This and also the equity and risk-sharing elements result in the absence of mismatch of short-term, on-sight demandable deposit contracts with long-term loan contracts. Speculation is not allowed and the banks are restricted from using many hedg-ing instruments which tends to increase the asset concentration (Beck et al. 2013) and it makes it more difficult for the banks to mitigate and diversify risk. Thus, it is not surprising that there is consistent evidence of higher capi-talization and higher liquidity reserves (Imam & Kpodar 2016). Moreover, the high geographical and sector concentration of Islamic banks further hinders the diversification.
The empirical results are ambiguous as well. For example, some authors find that Islamic banks are more cost and profit efficient than their conventional counterparts (Bashir 1999; Al-Jarrah & Molyneux 2010), some find no statis-tically significant difference (Majid 2003; Mohamad et al. 2008; Abdul Razak et al. 2008; Beck et al. 2013), others find that they are less efficient relative to the conventional banks (Majid et al. 2010; Johnes et al. 2009; Abdul-Majid et al. 2011; Johnes et al. 2014).
Chapter 3
Recent empirical studies
This chapter provides an outline of the recent empirical advancements. A lot of attention has been drawn on the performance comparison of the two types of banking. First of these types of studies are centred on single countries using different approaches from OLS regression and analysis of variance to stochastic frontier analysis. More recent studies tend to focus on a cross-country analysis, typically using parametric or non-parametric frontier modelling approaches to model cross-country bank cost, profit efficiency, or productivity (Abedifar et al. 2015).
One of the most detailed studies on the comparison of the two types of banking is the one by Beck et al. (2013). They investigate the differences in business orientation, efficiency, asset quality, and stability of Islamic and conventional banks using a sample of banks from 141 countries over the period from 1995 to 2007. They find few significant differences in business orientation. Nonetheless, they argue that Islamic banks are less cost-effective but have a higher intermediation ratio, higher asset quality and better capitalization.
Only recently have the scholars concentrated their attention to the link between Islamic banking development and its impact on the macroeconomic indicators. Gheeraert (2014) first pointed out that the development of Islamic banking could impact the economic growth through its effect on the overall banking sector. He argues that, firstly, it may increase the inclusion in the banking sector both by including previously “unbanked” individuals and by raising the volume deposited by existing clients. Moreover, the Islamic bank-ing development may spur competition and the need for innovation by its con-ventional counterparts, as new products become available in the market. If the banking market structure is indeed affected, two possible scenarios may
3. Recent empirical studies 10
emerge. Either will Islamic banks and windows possess too high market power, or oppositely, the competition will intensify and Islamic banks become more incentivized to lend money.
Gheeraert (2014) does not investigate the different channels separately, but assesses the overall net effect of Islamic banking on the overall banking sector development. As a proxy for this measurement he uses the total level of deposit or private credit relative to the economy. He finds that in countries with high Muslim population ratios, the level of the Islamic banking development strongly boosts the overall banking sector development. Given his dataset, this effect is estimated to be 7 % higher compared to the no Islamic banking case. He finds support for the hypothesis that Islamic banking serves as a complement to the conventional banking.
Gheeraert & Weill (2015) then come closer to the question concerning im-pact of Islamic banking development on economic growth. Numerous authors point out that the development of financial intermediation spurs economic growth. For example Levine (2005) argues that financial development can facilitate information availability, the enforcement, and can reduce transaction costs of financing decisions. Instead of concentrating on the economic growth, however, the authors decided to analyse the role of Islamic banking on produc-tivity and they justify this choice by two reasons. Firstly, scholars agree that productivity growth is more important than factor accumulation in explaining countries’ growth differences. And secondly, Islamic banking is quite recent phenomenon and analysing its long-term impact on economic growth might be premature.
Moreover though, due to the profit-and-loss sharing, Islamic banks have even stronger incentives to evaluate investment projects before the lending de-cision is made. They may therefore contribute to the production of ex ante information even better than conventional banks and thus stimulate the opti-mal allocation of capital. Another channel mentioned by the authors is that of increased participation in the formal banking system and thereby facilitated mobilization of resources from different economic agents. There is still the possibility, though, that these channels are not strong enough to create a com-parative advantage for Islamic banking relative to conventional one.
For the purpose of the analysis, Gheeraert & Weill (2015) use efficiency frontiers to aggregate production function . This method attempts to calculate countries’ relative distances from their estimated production frontier. That is, the greater production for a given bundle of inputs the country has, the
3. Recent empirical studies 11
more efficient it is. The production frontier is estimated through the stochastic frontier approach. Islamic banking development is measured by the ratios of credit and of the deposits of Islamic banks to the local GDP by country and by year. They find evidence that the development of Islamic banking indeed favours efficiency and the relationship is non-linear. However, this only holds up to the certain point of development and when this level is exceeded, detrimental effects on efficiency are possible.
Abedifar et al. (2016) attempt to fill in the gap in literature on the coex-istence of the two types of banking by investigating whether this coexcoex-istence can foster the development of the overall banking sector and economic welfare. This study of 22 Muslim countries with dual-banking system during 1999-2011 shows evidence that the market share of medium-size Islamic banks in mainly Muslim countries has a positive relationship with more efficient funds mobilization and private credit allocation, the two variables that proxy for the development of financial intermediation. Moreover, it is negatively linked to the income inequality as measured by the percentage of population below the national poverty line and headcount ratio in the rural area. Overall, the results support the hypothesis that the presence of Islamic banks can foster access to finance and financial deepening, and improve economic welfare in terms of lowering income inequality and alleviating the poverty.
Imam & Kpodar (2016) directly investigate whether Islamic banking is favourable for growth. They argue that Islamic banking has features that are likely to boost growth, however, this does not need to be projected to the macroeconomic level yet as the industry is still in the process of developing. For this reason they only examine whether Islamic banking development has positive impact on the economic growth. Despite many academics question the robustness of positive relationship between the development of financial sector and growth after certain level of development has been reached, espe-cially after the recent financial crisis, for developing countries the relationship is much clearer. There it boosts the growth especially through the acceleration of capital accumulation, while in rich countries finance stimulates the growth primarily by promoting productivity growth (Imam & Kpodar 2016).
Chapter 4
Empirical part
In this section I empirically test the effect of the development of Islamic bank-ing on the macroeconomic indicators. Literature emphasizbank-ing the importance of the financial sector development has been increasing in size. Accordingly, numerous authors show that measures of the size of the banking sector are highly correlated with the subsequent growth of GDP (Beck et al. 2000). I try to estimate the effect of the Islamic banking development on both. More specifically, based on the literature review the following two hypotheses will be tested:
Hypothesis 1:
Islamic banking development has a positive effect on the overall banking sector development.
Hypothesis 2:
Islamic banking development has a positive effect on macroeconomic growth.
4.1
Methodology and data description
I opt for panel data to analyse the hypotheses since they offer several benefits which improve the efficiency of the estimations. They can better control for individual heterogeneity providing more information, variability, and degrees of freedom, and on the other hand less collinearity between the variables (Baltagi 2008).
4. Empirical part 13
The cross-sectional dimension of my data is represented by majority of the countries that have more than 20% Muslim population. The only ex-ception is Thailand where it amounts only to 5.8 %. This small percentage, however, still translates to almost 4 million Muslims. That forms a sample of 49 countries. Out of that, 22 countries operate on the dual-banking sys-tem: Bahrain, Bangladesh, Brunei Darussalam, Egypt, Indonesia, Iraq, Jor-dan, Kuwait, Lebanon, Malaysia, Mauritania, Oman, Pakistan, Qatar, Saudi Arabia, Senegal, Syria, Thailand, Tunisia, Turkey, the United Arab Emirates, and Yemen. Gambia and Palestinian Territories operate on the dual-banking system as well, but are excluded from the sample as the data are not acces-sible. Islamic Republic of Iran and Sudan function on the basis of a fully Shari’ah-compliant financial system. The other 27 countries were included as the baseline countries to better explain the variations in the variables of our interest. These include Afghanistan, Albania, Algeria, Azerbaijan, Benin, Bosnia and Herzegovina, Burkina Faso, Cameroon, Chad, Comoros, The Gam-bia, Guinea-Bissau, Kazakhstan, Kosovo, the Kyrgyz Republic, the Republic of Macedonia, Maldives, Mali, Morocco, Mozambique, Niger, Nigeria, Sierra Leone, Somalia, Tajikistan, Tanzania, and Togo. Other countries fulfilling the minimum Muslim population share criterion were excluded due to the data unavailability.
The time-series dimension of my panel dataset comprises of years 2007 to 2014 with one observation per year. That gives a total of 8 years of observations for each country and 392 observations per each variable. My panel dataset is balanced with no observations missing.
Here I would like to stress the limitations of my dataset. Firstly, the sample is strongly selective. For that reason I attempted to include as many countries that correspond to my desired characteristics as possible to make the sample possibly most representative. Secondly, due to the data unavailability, I was able to extract only eight years of observations. That may hinder the analysis at the macroeconomic level where many influences display effects only in the long run. However, Islamic banking is a nascent industry and as it is growing at double digit rates after the recent financial crisis, its effects might be notable, especially on the banking sector development.
4. Empirical part 14
4.1.1
Methodology
To test the effect of the Islamic banking development on the two macroeconomic indicators of my interest I use the differences-in-differences design, i.e.:
Yit= β0+ β1Xit+ αi + λt+ Wit+ uit,
where i = 1, ..., N stands for the entity and t = 1, ..., T stands for the time period. Yit represents the dependent variable of entity i in time t, Xit
rep-resents the independent variable whose effect is being analysed, αi represents
the unobserved entity-fixed effects, i.e. the individual intercept for entity i, λtrepresents the unobserved time-fixed effects, Wit is the vector of covariates,
and uit is the error term.
This model attempts to simulate the experimental research design by com-paring the differential effect of a “treatment” on the treatment group compared to the control group (Angrist & Pischke 2008). In more detail, the impact of Islamic banking development on the overall banking sector development will be estimated by comparing the average change in banking sector development in countries where Islamic banking is established (controlling for different levels of penetration) to that in the countries where it is not instituted. Similarly for the GDP growth.
For the purpose of differences-in-differences estimation it is convenient to have the data before, during and after the treatment has been applied, to be able to extract the pure effect of the treatment. Instead, in this case the “treat-ment” can be understood of as the presence of Islamic banking in that country’s banking sector and for different levels of the Islamic banking penetration the outcome variable is compared to the countries with the same characteristics where Islamic banking is not established.
One of the main assumptions underlying any econometric model is the zero conditional mean. Specifically for panel data the error term composes of two parts:
eit = uit+ αi.
αi represents the unobserved entity-specific component which is time-invariant,
and uitrepresents the component that varies both over time and across entities.
The panel data estimation techniques assume that the error term has fixed or random one-way error structure. This structure depends on whether the un-observable individual-specific factors are correlated with the regressors. While
4. Empirical part 15
random-effect model postulates that this correlation is zero, the fixed-effects model leaves that correlation unrestricted. As I analyse quite homogenous group of countries, fixed-effects estimation technique is preferred as opposed to a case with randomly composed sample.
To assess the specification decision the Hausman’s (1978) specification test is applied. In this case, it compares the fixed-effects estimator that is assumed to be consistent with the random-effects estimator that is efficient under the assumption being tested. Under the null hypothesis, the random-effects is indeed an efficient and consistent estimator of the true parameters. If the estimates are systematically different between the two estimators, though, i.e. the null hypothesis is rejected, then the random-effects estimator is not efficient and should not be used (StataCorp. 2011). The Hausman’s test supports the decision to use the fixed-effects estimator.
The component αi may embody any country-specific characteristics that is
not observed. It captures the heterogeneity across the countries in the variables that do not change over time. The fixed-effects estimation technique eliminates this characteristics’ effect by “within transformation” which demeans all the variables. I.e., for a simple panel data equation:
Yit= β1Xit+ uit,
where Yit is the dependent variable for entity i in time t, Xit the independent
variable and uit the error term, firstly the arithmetic means of the variables
(denoted with over bars) and then the deviations from corresponding means are computed:
¯
Yit = β1X¯it+ ¯αi+ ¯uit,
(Yit− ¯Yit) = β1(Xit− ¯Xit) + (αi− ¯αi) + (uit− ¯uit).
The fixed-effect estimator bβ1is then obtained by the following ordinary least
squares regression:
¨
Yit= β1X¨it+ ¨uit,
where ¨Yit = (Yit − ¯Yit), X¨it = (Xit − ¯Xit), and ¨uit = (uit − ¯uit). Because
the country-specific effect is time-invariant, it is thereby eliminated from the equation, i.e. the “within” component of the error term does not have to be taken into account anymore. The essential intuition behind the entity-fixed
4. Empirical part 16
effects is that if some characteristics of the entity are invariant over time, any changes in the outcome variable must be due to the effects of other variables (Stock & Watson 2003).
Moreover, due to the similarity of the countries in my sample across several aspects, I account for the time-fixed effects by including the time dummy vari-ables in the equation. These control for the influences that are identical across entities but vary over time.
Because αi is removed, it cannot be a source of serial correlation. However,
uit might still be serially correlated and heteroskedastic. And indeed, the test
for groupwise heteroskedasticity proposed by Greene et al. (2001) suggests that the null hypothesis of homoskedasticity is rejected. The test, however, has a very low power in small samples, especially with large N and small T , and should thus be taken with caution. Regarding the serial correlation, to my best knowledge there is not a test developed that would examine its presence in the error term in fixed-effects models with the time-fixed effects included. It is not unlikely, though, that the serial correlation is present, especially in regressions with the overall banking sector development as a dependent variable.
Therefore I opt to use the “cluster-robust” inference. These heteroskedastic-and autocorrelation-robust stheteroskedastic-andard errors are robust to arbitrary correlation within clusters (countries), assuming no correlation across entities.
Besides the assumptions touched upon above, others are crucial for the ap-propriate inference as well. These are that (Xi1, . . . , XiT, ui1, . . . , uiT) are
inde-pendent and identically distributed over n, the cross-section. Further, it needs to be assumed that large outliers are unlikely and no perfect multicollinearity is present. Differences-in-differences model uses the trends to form the counter-factual. Therefore, an additional assumption has to hold, the one of a parallel trend across entities. Given these assumptions hold, the fixed-effect estimator in this differences-in-differences model is unbiased and consistent.
Literature stressing the issue of cross-sectional dependence in panel data is growing in size. Recently, we have experienced an ever-increasing economic and financial integration of countries. It is not unlikely that the panels in the dataset are interdependent due to the very similar country characteristics and neighbourhood effects, etc. The impact of this dependence depends on its magnitude and its nature. Assuming the cross-sectional dependence is caused by common unobserved factors which are uncorrelated with the included ex-planatory variables, the fixed-effects estimator is consistent, but not efficient and the estimated standard errors are biased. However, if these factors are
4. Empirical part 17
correlated with the included explanatory variables, the fixed-effect estimator will be biased and inconsistent (De Hoyos et al. 2006).
The dataset contains more entities than years and therefore the Lagrange multiplier test conditions are not satisfied and the test cannot be performed. Therefore I follow the semi-parametric tests proposed by Friedman (1937) and Frees (1995; 2004) and parametric test proposed by Pesaran (2004) that are valid when T < N .
These tests examine the hypothesis of cross-sectional independence and in this case they do not reject it. Thus, cross-sectional dependence should not be an issue in my data. However, these tests are designed for small T and large N and N of 49 need not be large enough, therefore the inference should be taken with caution.
Given the macroeconomic nature of my data, the large outliers that would extremely drive the results are unlikely. Another issue to deal with is the potential multicollinearity. I first investigate the correlation matrix and con-clude that there is not too high correlation between any of the two variables and thus none has to be excluded. Furthermore, I compare the values of the variance inflation factor to a benchmark equal to ten, which is considered as a value signing the multicollinearity problem, and confirm that a perceptible multicollinearity is not present.
To assess the validity of my design I will investigate the effect of Islamic banking development in several steps. First I will estimate the following re-duced forms so that the outcomes of the different specifications can be compared and examined:
Yit = β0+ β1Xit+ αi+ uit, f or i = 1, ..., N, t = 1, ..., T, (4.1)
Yit = β0 + β1Xit+ αi+ λt+ uit, f or i = 1, ..., N, t = 1, ..., T. (4.2)
In the Equation 4.1 only entity-fixed effects are assumed to be present. Including the time-fixed effects, as in Equation 4.2, can change the estimates given the time-fixed effects are present, as the former equation would in that case suffer the omitted-variable bias. Given that the effect of the independent variable is exogenous to the other covariates, whether they are included or not should not affect the estimation output regarding the independent variable.
Only then will I estimate the third specification incorporating the covariates: Yit= β0+ β1Xit+ αi+ λt+ Wit+ uit, f or i = 1, ..., N, t = 1, ..., T. (4.3)
4. Empirical part 18
4.1.2
Description of variables
Dependent variables
Banking sector development As I want to test the effect of the development of the Islamic banking on the overall banking sector development, I need an appropriate measure of the latter. I use the traditional indicator utilized mainly for assessing the size of the banking sector, the total banking deposits relative to the size of economy, i.e. to the GDP, a variable named bdgdp. I retrieve the “bank deposits to GDP” data from the Financial Development and Structure Dataset of the Financial Structure Database, The World Bank.
GDP growth The dependent variable in the second set of estimations is the economic growth (annual growth rate of GDP) retrieved from the World Devel-opment Indicators of The World Bank. More specifically, the annul percentage growth rate of GDP is measured at market prices based on the local currency, aggregates are based on constant 2010 USD. It is denoted GDP growth. Independent variable
Islamic banking development The regressor of my interest is the Islamic banking development, approximated by the measure of penetration of Islamic deposits in the overall banking sector within the country, i.e. the share of Islamic deposits in the overall banking sector deposits. The data on Islamic banking deposits were retrieved from the “deposits and short-term funding” of Orbis Bank Focus, the most detailed and contemporary financial format available. However, even this database is not exhaustive and does not contain data on all banks across the relevant countries. For a more precise measure I looked up the missing data on the Shari’ah-compliant deposits in the annual reports of all banks reported in the dataset as Islamic. Accordingly, I do not have missing observations for the depicted banks’ deposits through 2011 to 2014. From 2007 to 2010, however, the data were available not in all cases. I will use the narrower sample with no missing deposits information for a robustness check later on. Moreover, the Orbis Bank Focus dataset does not contain information about the Islamic windows and these are thus not included in the dataset. All in all, the Shari’ah-compliant deposits are understated but hopefully not significantly.
Because I use the ratio of Islamic deposits to the overall banking deposits, I need the observations on the latter. This was obtained by multiplying the
4. Empirical part 19
bank deposits to GDP ratio by the GDP in constant 2010 USD, retrieved from The World Development Indicators of The World Bank.
Figure 4.1: Histogram of the penetration level of Islamic banking in the overall banking sector
0 5 10 15 Density 0 .2 .4 .6 .8 1
Share of Islamic deposits in the overall banking sector deposits (isl_bd)
Source: Graph based on author’s computations in Stata 12.0.
Covariates
As mentioned above, the development of the banking sector and the GDP growth are theoretically highly correlated (Beck et al. 2000). Therefore I use the same set of covariates. The variables were retrieved from the World Bank, if not stated otherwise.
Index of the strength of legal rights Variable legal index represents the strength-of-legal-rights index. “It measures the degree to which collateral and thus bankruptcy laws protect the rights of borrowers and lenders and thus facilitate lending. The index ranges from 0 to 12, with higher scores indicating that these laws are better designed to expand access to credit,” as stated in the World Bank data description. As follows, the expected effect on the banking sector development as well as on the GDP growth is positive.
4. Empirical part 20
Figure 4.2: Cumulative distribution function of the penetration level of Islamic banking in the overall bankung sector
0 .5 1 0 .5 1 0 .5 1 0 .5 1 0 .5 1 0 .5 1 2007 2008 2009 2010 2011 2012 2013 2014
Cumulative distribution function of isl_bd
Share of Islamic deposits in the overall banking sector deposits (isl_bd)
Graphs by year
Source: Graph based on author’s computations in Stata 12.0.
Inflation Variable inflation as measured by the consumer price index “reflects the annual percentage change in the cost to the average consumer of acquiring a basket of goods and services that may be fixed or changed at specified intervals, such as yearly. The Laspeyers formula is generally used,” as stated in the World Bank data description. The higher the inflation, the higher uncertainty it brings to the financial system and its main processes might be impeded. Thus, the level of inflation and the banking sector development should be negatively correlated. As follows from the high interconnectedness of the banking sector development and GDP growth, the relationship between the level of inflation and GDP growth is expected to be negative as well.
Quality of institutions Variable corruption is supposed to serve as a proxy for the quality of institutions. The “control of corruption estimate” retrieved from the World Bank attempts to capture the “perceptions of the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as capture of the state by elites and private interests,” as stated in the World Bank data description. It ranges from -2.5
4. Empirical part 21
to 2.5 with higher values representing worse institutions as they are subject to higher levels of corruption. The worse the quality of institutions, the lower the banking sector development should be and the same applies for the GDP growth. The expected coefficient is thus negative.
Trade openness Variable trade stands for the sum of exports and imports of goods and services measured as a share of GDP in %. The higher trade openness may boost the GDP growth and banking sector development due to, among other things, the technological advancements it brings into the economy and the higher economic activity it may be correlated with. On the other hand though, the economy might be more vulnerable to external shocks when it is more open for outside trade. The expected relationship is thus ambiguous. Political-stability index The variable polit index was retrieved from theglob-aleconomy.com. It is a composite measure and is based on several other indexes from multiple sources. “The underlying indexes reflect the likelihood of a dis-orderly transfer of government power, armed conflict, violent demonstrations, social unrest, international tensions, terrorism, as well as ethnic, religious or regional conflicts” . The index ranges from -2.5 to 2.5 indicating weak to strong stability, respectively. The expected relationship with the dependent variables is clearly positive.
Government expenditures The variable L gov exp perc represents the gen-eral government final consumption expenditure measured as % of GDP. I as-sume it has (positive) effect on the subsequent GDP growth and thereby on the banking sector development, therefore I use the lagged values.
Population growth The last control variable included is the twice lagged population growth in %, L2 pop growth. I assume that the effect of population growth does not have an immediate influence and therefore the two lags. The lower fertility has been shown to have positive effect on GDP growth (Barro 1996) and therefore the expected relationship is negative.
Other covariates were also taken into consideration but eventually were not included in the equations. One of the most relevant is a measure of the level of education. Since, to my best knowledge, only the data on the primary, secondary, and tertiary education enrolment rate are available and not those
4. Empirical part 22
indicating the level of the education of the productive-age individuals, I decided to exclude them as irrelevant. The effect of enrolment rate is likely to display some effect when these individuals start to work, several years later. Omitting this, however, may not matter in such short sample, as I assume that education level does not change significantly over 8 years. It can be thought of as a part of the entity-fixed effects.
The overview and descriptive statistics of the variables in the dataset are presented in Table 4.1.
The equations to be estimated for the first hypothesis thus become:
bdgdpit = β0+ β1isl bdit+ αi+ uit, (4.4)
bdgdpit= β0+ β1isl bdit+ αi+ λt+ uit, (4.5)
bdgdpit = β0+ β1isl bdit+ αi+ λt+ Wit+ uit, (4.6)
for i = 1, . . . , N , t = 1, . . . , T , and the equations to be estimated for the second hypothesis:
GDP growthit= β0+ β1isl bdit+ αi+ uit, (4.7)
GDP growthit = β0+ β1isl bdit+ αi+ λt+ uit, (4.8)
GDP growthit= β0+ β1isl bdit+ αi+ λt+ Wit+ uit, (4.9)
where Wit is the varying vector of covariates. The first variation of the Wit is
the one where government expenditures and population growth are excluded. The reason for this is that I do not expect the direct effect of these variables on the banking sector development. And to make the analysis of the two different hypotheses consistent, I decided to use the same variations of this covariates vector in both. The second variation is the one where all the explained con-trol variables are included. In the last estimation, the variable expressing the political-stability index is excluded due to its pronounced insignificance.
4.2
Empirical results
4.2.1
Hypothesis 1
I begin with the empirical analysis of the first proposed hypothesis:
Islamic banking development has a positive effect on the overall banking sector development.
4. Empirical part 23
Table 4.1: Summary statistics
T yp e of the v aria ble Name of the v ariable Description of the v a riable Descriptiv e statistics Mean Std. Dev. Min Max Dep enden t v aria bles b dgdp Ratio of ban k dep osits to GDP in % 41.98 36.16 4.47 2 45.43 GDP gro wth GDP gro wth in % 4.28 3.61 -12.71 25.05 Indep enden t v aria ble isl b d Share of Islamic dep osits in the o v erall banking sector dep osits 0.08 0.21 0.00 1.00 legal index Strength of legal righ ts index (0=w eak to 12=strong) 3.90 2.45 0.00 9.00 inflation Inflation, consumer pric es (ann ual %) 6.16 6.70 -10.07 39.27 corruption Corruption index, pro xy fo r the qualit y of institutions (-2.5 strong, 2.5 w eak) -0.54 0.62 -1.64 1.72 Co v ariates trad e Sum of exp orts and imp orts of go o ds and services measured as a share of GDP in % 81.18 35.60 19.46 204.59 p olit index P olitical stabilit y index (-2.5 w eak, 2.5 strong) -0.67 0.95 -2.81 1.36 L go v exp p erc General go v ernmen t final consumption exp enditure (% of GDP), lagged b y one y ear 14.66 4.72 4.57 27.97 L2 p op gro wth P opulation gro wth (ann ual %), lagged b y tw o y ears 2.66 2.51 -2.58 17.62
4. Empirical part 24
Table 4.2 depicts estimation results. The first and second regression output displayed are the results of estimations of Equation 4.4 and Equation 4.5, re-spectively. The output of the third, fourth, and fifth regressions are variations of Equation 4.6 where different covariates are taken into account.
Including the time-fixed effects changes the regression output notably. The statistical significance at 1% level of all the time dummy variables indicates that the time-fixed effects are present. Testing for their joint significance gives supportive evidence with an F statistics of 16.91. Moreover, R2 increases from
0 to 0.35.
Controlling for different covariates possibly affecting the bdgdp variable does not alter the inference about isl bd much, which is in line with our expectations. More specifically, including them into the regression changes the estimated coefficient on isl bd from -29.44 to -25.34, -25.2, and −25.08 in (3), (4), and (5), respectivelly. That would mean that each 10% (or equivalently 0.1 unit) increase in the Islamic banking penetration level in the overall banking sector over some period of time is associated with a 2.52% higher difference between the countries with and without Islamic banking in the ratio of bank deposits to GDP, given the only difference between the two countries is the presence of Islamic banking. More specifically, the higher the Islamic banking penetration level, the lower the ratio of bank deposits to GDP that such country would have. The 95% confidence interval, given the assumptions of the model hold and the standard errors are not biased and inconsistent, ranges approximately from -49.1 to -4.7, combining (3), (4), and (5). Given the high pace of growth of Islamic banking in the recent years and the positive prospects of the subsequent growth, that is not a negligible impact on the banking sector development (or at least on its volume). Especially when we realize that the mean of the proxy for the banking sector development is 42 % in my sample. As I wanted to investigate whether the development of Islamic banking has positive effect on the overall banking sector development, my empirical results with significantly negative estimated effect contradict the proposed hypothesis.
Out of the incorporated covariates only two are significant at 5% level. A higher index of legal rights by 1 (where the scale is from 0 to 12) is on average associated with 4% higher bank deposits to GDP ratio. Similarly, a higher volume of trade scaled by GDP by 10 % is associated with on average 1% higher bank deposits to GDP ratio. Furthermore, the t statistics of the estimated coefficient on inflation is ranging from 1.59 to 1.75. Together, nonetheless, all the control variables are statistically significant with an F statistics of 7.03,
4. Empirical part 25
5.20, and 6.03 in (3), (4), and (5), respectively. The R2 and adjusted R2
increase as well with the inclusion of the covariates.
Table 4.2: Results of the first hypothesis tests.
The table displays the fixed-effects estimation results of Equation 4.4, Equa-tion 4.5, and EquaEqua-tion 4.6.
Equation 4.4 Equation 4.5 Equation 4.6
(1) (2) (3) (4) (5) bdgdp bdgdp bdgdp bdgdp bdgdp isl bd 4.645 -29.44∗∗∗ -25.34∗∗ -25.20∗∗ -25.08∗∗ (9.274) (9.766) (10.19) (9.875) (10.10) 2007.year 0 0 0 0 (.) (.) (.) (.) 2008.year 1.233∗∗∗ 0.633 0.491 0.484 (0.355) (0.501) (0.507) (0.527) 2009.year 4.617∗∗∗ 5.369∗∗∗ 5.378∗∗∗ 5.372∗∗∗ (1.021) (0.964) (0.984) (0.988) 2010.year 4.920∗∗∗ 5.465∗∗∗ 5.238∗∗∗ 5.224∗∗∗ (0.946) (0.851) (0.867) (0.867) 2011.year 5.449∗∗∗ 5.415∗∗∗ 5.288∗∗∗ 5.265∗∗∗ (0.881) (0.799) (0.768) (0.807) 2012.year 6.619∗∗∗ 6.313∗∗∗ 6.198∗∗∗ 6.173∗∗∗ (1.061) (0.960) (0.951) (0.978) 2013.year 7.842∗∗∗ 7.684∗∗∗ 7.550∗∗∗ 7.526∗∗∗ (1.189) (0.989) (0.970) (1.034) 2014.year 9.712∗∗∗ 9.920∗∗∗ 9.720∗∗∗ 9.702∗∗∗ (1.316) (1.050) (1.030) (1.084) legal index 4.018∗∗∗ 4.006∗∗∗ 4.004∗∗∗ (0.824) (0.840) (0.845) inflation 0.0574 0.0717∗ 0.0724∗
4. Empirical part 26 (0.0362) (0.0409) (0.0421) corruption 2.337 2.102 2.168 (2.688) (2.682) (2.559) trade 0.104∗∗ 0.101∗∗ 0.101∗∗ (0.0499) (0.0488) (0.0485) polit index -0.00107 0.0966 (1.232) (1.190)
L gov exp perc 0.173 0.173
(0.144) (0.143) L2 pop growth -0.119 -0.116 (0.361) (0.362) cons 41.62∗∗∗ 39.18∗∗∗ 15.62∗∗∗ 13.63∗∗ 13.61∗∗ (0.708) (0.811) (5.541) (5.897) (5.933) N 392 392 392 392 392 R2 0.000 0.350 0.425 0.430 0.429 adj. R2 -0.002 0.336 0.406 0.407 0.408
Clustered standard errors in parentheses,
∗ p < .10,∗∗ p < .05,∗∗∗ p < .01
Source: Author’s computations in Stata 12.0.
4.2.2
Hypothesis 2
I proceed with the empirical analysis of the second proposed hypothesis: Islamic banking development has a positive effect on macroeconomic growth. The regression tests outputs are displayed in Table 4.3. Similarly as before, the dummy variables capturing the time-fixed effects are statistically significant at 1% level each. Their joint significance is supported by the F statistics of 7.06. Moreover, R2 increases from 0 to 0.106 when they are included.
According to my estimations, Islamic banking does not have significant influence on the GDP growth. For example, given the output of regression (5) in Table 4.3, the coefficient on isl bd falls within -4.2 and 29 with 95% probability in case the estimates are unbiased and consistent. That is not
4. Empirical part 27
statistically significantly different from zero. Therefore I cannot reject that the hypothesis of positive effect of the Islamic banking development on the GDP growth is wrong.
Most of the variability in the GDP growth is captured by the most signifi-cant year dummies. Except for them, variable representing trade volume has a statistically significant positive effect on the GDP growth. An increase in the trade to GDP ratio by 10 % is associated on average with a 0.56% higher GDP growth. Also 1% higher previous-year government expenditures are on average followed by around 0.3% higher GDP growth. The other variables incorporated are individually not significant. Together, however, the control variables have statistically significant effect on the GDP growth with the F statistics of 2.48, 2.34, and 2.45 for (3), (4), and (5), respectively.
Table 4.3: Results of the second hypothesis tests.
The table displays the fixed-effects estimation results of Equation 4.7, Equa-tion 4.8, and EquaEqua-tion 4.9.
Equation 4.7 Equation 4.8 Equation 4.9
(1) (2) (3) (4) (5)
GDP GDP GDP GDP GDP
growth growth growth growth growth
isl bd 1.127 7.857 12.06 11.85 12.37 (10.89) (12.02) (9.753) (8.151) (8.259) 2007.year 0 0 0 0 (.) (.) (.) (.) 2008.year -1.204∗∗ -1.038 -1.274∗ -1.302∗ (0.532) (0.715) (0.699) (0.684) 2009.year -3.374∗∗∗ -3.020∗∗∗ -2.992∗∗∗ -3.020∗∗∗ (0.678) (0.617) (0.590) (0.579) 2010.year -1.116 -0.813 -1.190∗ -1.255∗ (0.685) (0.690) (0.709) (0.672) 2011.year -2.941∗∗∗ -2.775∗∗∗ -2.984∗∗∗ -3.087∗∗∗ (0.836) (0.868) (0.897) (0.847) 2012.year -2.065∗∗ -1.973∗∗ -2.161∗∗ -2.277∗∗
4. Empirical part 28 (0.834) (0.927) (0.921) (0.852) 2013.year -2.122∗∗∗ -2.027∗∗∗ -2.253∗∗∗ -2.364∗∗∗ (0.674) (0.754) (0.751) (0.685) 2014.year -2.725∗∗∗ -2.535∗∗∗ -2.879∗∗∗ -2.958∗∗∗ (0.692) (0.753) (0.791) (0.746) legal index 0.166 0.155 0.147 (0.174) (0.171) (0.172) inflation -0.0589 -0.0343 -0.0311 (0.0901) (0.0867) (0.0871) corruption 0.893 0.446 0.740 (1.040) (0.915) (0.948) trade 0.0609∗∗ 0.0562∗∗ 0.0558∗∗ (0.0245) (0.0234) (0.0234) polit index 0.246 0.436 (0.476) (0.450)
L gov exp perc 0.297∗∗∗ 0.295∗∗∗
(0.0899) (0.0904) L2 pop growth -0.249 -0.233 (0.171) (0.167) cons 4.189∗∗∗ 5.618∗∗∗ 0.549 -2.773 -2.848 (0.832) (0.937) (2.625) (2.670) (2.672) N 392 392 392 392 392 R2 0.000 0.106 0.143 0.175 0.174 adj. R2 -0.002 0.087 0.114 0.142 0.143
Clustered standard errors in parentheses,
∗ p < .10,∗∗ p < .05,∗∗∗ p < .01
4. Empirical part 29
4.3
Robustness checks
In this part I will apply the robustness checks on my analysis. Firstly, I will exclude Lebanon from the first hypothesis testing as its bank deposits to GDP ratio is much higher than in the other countries over the whole period and that could possibly drive the results. The output of these regressions is presented in Appendix A. The estimated effect of Islamic banking development on the banking sector development has not changed significantly by this inclusion. The estimated coefficient is still close to -25.
As a second robustness check the sample is cut to only 2011 till 2014. The decision to analyse this narrower sample follows from the data collection. In the first years of the sample some annual reports with the information about deposits were not available, resulting in understatement of the Islamic deposits in these years, as opposed to the 2011 to 2014 where all were available. As follows, the growth of Islamic deposits is then overstated in the sample and that may lead to an overstatement of its effect on the two analysed variables. Nonetheless, it has to be borne in mind that this procedure results in an even smaller sample making the estimation more complicated.
The results of the second robustness check are displayed in Table 4.4 and Table 4.5 for the first and second hypothesis, respectively. The effect of Is-lamic banking development on the overall banking sector development turned insignificant. The results for the second hypothesis do not change when consid-ering shorter period of time. The effect is still insignificant and the estimated coefficient is similar to the one in the original analysis (e.g. in regression (5) now it is 8.9 vs. 12.3).
Table 4.4: Results of the second robustness check for banking sector development: 2011–2014.
The table displays the fixed-effects estimation results of Equation 4.4, Equa-tion 4.5, and EquaEqua-tion 4.6.
Equation 4.4 Equation 4.5 Equation 4.6
(1) (2) (3) (4) (5)
bdgdp bdgdp bdgdp bdgdp bdgdp
isl bd -7.786 -15.72 -11.83 0.733 -1.031
(4.772) (12.97) (10.18) (13.46) (14.05)
4. Empirical part 30 (.) (.) (.) (.) 2012.year 1.114∗∗ 0.792∗ 0.773∗ 0.733∗ (0.446) (0.404) (0.415) (0.418) 2013.year 2.373∗∗∗ 2.052∗∗∗ 1.943∗∗∗ 1.903∗∗∗ (0.606) (0.591) (0.542) (0.547) 2014.year 4.204∗∗∗ 3.980∗∗∗ 3.526∗∗∗ 3.619∗∗∗ (0.726) (0.640) (0.576) (0.589) legal index 4.113∗∗∗ 4.134∗∗∗ 4.066∗∗∗ (0.873) (0.872) (0.962) inflation -0.0641 -0.0703 -0.0682 (0.0745) (0.0656) (0.0685) corruption -2.010 -3.085 -1.917 (2.851) (2.943) (2.796) trade 0.0305 0.0274 0.0193 (0.0348) (0.0339) (0.0328) polit index 0.564 1.766 (1.223) (1.186)
L gov exp perc 0.567∗∗ 0.493∗
(0.253) (0.256) L2 pop growth -0.521 -0.459 (0.314) (0.314) cons 44.78∗∗∗ 43.52∗∗∗ 24.50∗∗∗ 17.08∗∗∗ 18.38∗∗∗ (0.397) (1.066) (4.782) (6.156) (6.498) N 196 196 196 196 196 R2 0.002 0.289 0.373 0.428 0.417 adj. R2 -0.004 0.274 0.343 0.393 0.386
Clustered standard errors in parentheses,
4. Empirical part 31
Source: Author’s computations in Stata 12.0.
Table 4.5: Results of the second robustness check for the GDP growth: 2011–2014.
The table displays the fixed-effects estimation results of Equation 4.7, Equa-tion 4.8, and EquaEqua-tion 4.9.
Equation 4.7 Equation 4.8 Equation 4.9
(1) (2) (3) (4) (5)
GDP GDP GDP GDP GDP
growth growth growth growth growth
isl bd 4.114 2.959 3.651 9.908 8.914 (13.31) (13.83) (15.03) (16.61) (16.29) 2011.year 0 0 0 0 (.) (.) (.) (.) 2012.year 0.896 0.898 0.958 0.936 (0.797) (0.805) (0.792) (0.795) 2013.year 0.826 0.642 0.673 0.651 (0.729) (0.662) (0.621) (0.628) 2014.year 0.237 -0.145 -0.295 -0.242 (0.616) (0.623) (0.634) (0.618) legal index 0.154 0.142 0.103 (0.252) (0.243) (0.233) inflation -0.123 -0.104 -0.103 (0.140) (0.137) (0.138) corruption 2.984 2.374 3.033 (2.053) (1.963) (1.880) trade -0.0186 -0.0291 -0.0337 (0.0464) (0.0482) (0.0466) polit index 0.556 0.996 (1.102) (1.147)
4. Empirical part 32
L gov exp perc 0.474∗ 0.432
(0.255) (0.267) L2 pop growth 0.194 0.229 (0.259) (0.242) cons 3.466∗∗∗ 3.072∗∗ 6.900 -0.311 0.424 (1.106) (1.282) (4.254) (4.769) (4.775) N 196 196 196 196 196 R2 0.001 0.021 0.057 0.088 0.084 adj. R2 -0.005 0.001 0.011 0.034 0.035
Clustered standard errors in parentheses,
∗ p < .10,∗∗ p < .05,∗∗∗ p < .01
Source: Author’s computations in Stata 12.0.
To assess which results are more plausible, we have to weigh the benefits and costs of each analysis. In the original one, the Islamic banking deposits are understated during the first years. On the other hand, it gives three more years of observations and that is undoubtedly a large benefit.
4.4
Discussion of results
The results from section 4.2 show that Islamic banking development seems to have negative effect on the overall banking sector development. This could be attributed to the fact that Islamic banking is still in the early years of its devel-opment and the lack of proper accounting standards, cleaning and settlement processes may result in less transparency. That might hinder the overall bank-ing sector development rather than boost it (Imam & Kpodar 2016). Further, for the Islamic banking is still very small in size in many countries, it lacks the benefits of economies of scale. Moreover, if the Islamic banking functions as substitute rather than complement to the conventional banking, the overall development may drop as the less developed Islamic banks seize higher share. These issues should become less limiting as the industry matures.
There is also a possibility, though, that its negative effect on the overall banking sector development is not dependent on its maturity, but rather on the systematic flaws it might possess. It could be that it performs poorly in
4. Empirical part 33
conducting the main banking activities such as savings mobilization or alloca-tion of resources. Moreover, as discussed in the chapter 2 part, the ban to use derivatives and other financial instruments makes it more complicated to mit-igate and diversify risk. Due to the lack of liquid instruments, Islamic banks are forced to have large liquidity buffers putting them into disadvantageous position compared to the conventional banks (Imam & Kpodar 2016).
The insignificant effect on the GDP growth is less of a surprise. The GDP growth of each country is determined by manifold factors of which many are interconnected. It is thus very challenging to separate and determine the indi-vidual effects. Especially then for the industry that is still relatively small in many of the countries. It is also likely that the GDP growth is more related to the overall banking sector development regardless how the sector is constituted. Moreover, the small sample size limits the inference credibility which is further aggravated by the generally challenging modelling of the macroeco-nomic indicators due to the possible endogeneity problem, such as simultaneous causality. That has to be kept in mind when drawing conclusions. As the in-dustry evolves, though, the accessibility to the data on Islamic banking should improve and that should facilitate the analysis. The turn from significant to insignificant effect of Islamic banking development on the overall banking sec-tor development when applying the robustness check of a shorter period casts more doubts on my estimates. However, that might be simply due to the even smaller sample size.
Last but not least, one could question the appropriateness of the presented banking sector development measure. The deposits to GDP ratio captures only one aspect of the development, namely its size. However, other aspects might be as well so relevant. These would include for example M2 to GDP ratio capturing the degree of monetization or the private credit to GDP ratio expressing its functionality as an intermediary. Further, one might be interested in the measures of access to the financial system, including the branch and ATM density, household access, or average loan and deposits size. The quality of the financial system, on the other hand, can be captured by the measures of efficiency and stability (The World Bank 2006). Even if the development of Islamic banking influences negatively one of the aspects, it may work in the other direction on the other aspects. The investigation of the relationship of Islamic banking development to the other measures of the banking sector development might be object of further research.
4. Empirical part 34
the banking sectors of the countries with Muslim population. However, it has not yet been extensively examined by the researchers. Given its potential and space for improvement at the same time, it deserved much more attention. Since the results are often ambiguous within this field, it is important to provide more out-of-sample estimations and new analytical approaches.
Chapter 5
Conclusion
Islamic banking is a relatively new phenomenon that has received an ever-increasing attention by the researchers. That is not surprising given its growth pace and given how understudied it is at the same time. After the recent financial crisis, the Islamic banking assets have grown at double-digit rates and are expected to continue in this trend in the following years (EY 2016). As the Islamic banking is at the early stages of its development, the consensus regarding its overall impact on the macroeconomic indicators has not yet been agreed upon. The results should be, however, very important for the policy makers, since their decisions can influence its future notably. Being sure that it has uplifting impact would force the policy makers to adopt decisions that create a favourable setup for the further advancement.
Therefore I decided to contribute to the existing literature by another econo-metric approach, specifically the differences-in-differences analysis of the effect of Islamic banking development on the two important indicators of the eco-nomic prosperity. As many authors suggested and found, the banking sector development is closely correlated with the subsequent GDP growth (Beck et al. 2000). As Schumpeter & Backhaus (2003) argue, the financial development has a crucial role in economic development because the financial sector decides on the reallocation of resources. Since these two are highly correlated, I decided to test the effect of Islamic banking development on both in two different sets of estimations.
The Islamic banking development is approximated by its penetration level in the overall banking sector. The development of banking sector is proxied by the standard measure of its size, the bank deposits to GDP ratio. In each set of estimations I first investigate the regression with only entity-fixed effects, then
