The recovery of the Islamic and conventional banking systems from the 2008 financial crisis : an analysis on GCC (2006-2015)

Faculty of Economics and Business

Amsterdam School of Economics

Requirements thesis MSc in Econometrics.

1. The thesis should have the nature of a scientic paper. Consequently the thesis is divided up into a number of sections and contains references. An outline can be something like (this is an example for an empirical thesis, for a theoretical thesis have a look at a relevant paper from the literature):

(a) Front page (requirements see below)

(b) Statement of originality (compulsary, separate page) (c) Introduction (d) Theoretical background (e) Model (f) Data (g) Empirical Analysis (h) Conclusions

(i) References (compulsary)

If preferred you can change the number and order of the sections (but the order you use should be logical) and the heading of the sections. You have a free choice how to list your references but be consistent. References in the text should contain the names of the authors and the year of publication. E.g. Heckman and McFadden (2013). In the case of three or more authors: list all names and year of publication in case of the rst reference and use the rst name and et al and year of publication for the other references. Provide page numbers.

2. As a guideline, the thesis usually contains 25-40 pages using a normal page format. All that actually matters is that your supervisor agrees with your thesis.

3. The front page should contain:

(a) The logo of the UvA, a reference to the Amsterdam School of Economics and the Faculty as in the heading of this document. This combination is provided on Blackboard (in MSc Econometrics Theses & Presentations).

(b) The title of the thesis

(c) Your name and student number (d) Date of submission nal version

(e) MSc in Econometrics

(f) Your track of the MSc in Econometrics 1

U

NIVERSITY OF

A

MSTERDAM

Section Quantitative Economics

MS

T

E

F

T

The recovery of the Islamic and conventional banking

systems from the 2008 financial crisis:

An analysis on GCC (2006-2015)

Author:

A.G.M. (Sander) Cremers 11111615 Supervisor: dr. J.C.M. van Ophem Associate professor Company: KPMG Nederland

Financial Risk Management

Second reader: dr. S.A. Broda Assistant Professor

Date of submission: January 30, 2017

i

Statement of Originality

This document is written by Sander Cremers who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

ii

Abstract

in this thesis, we investigate the recovery of the Islamic and conventional banking sys-tems in the GCC region from the 2008 financial crisis based on fifteen Islamic banks and fifteen conventional banks. This is done using performance (profitability, risk, liquid-ity and asset-qualliquid-ity) and efficiency (cost and profit) as indicators. Regarding perfor-mance, the Islamic banks in the sample recovered better in terms of asset-quality, while the conventional banks recovered better in terms of profitability and risk. The results on liquidity are inconclusive. Regarding efficiency, we use both nonparametric data envel-opment analysis and parametric stochastic frontier analysis. For the parametric model, several functional forms and ways to deal with negative profits are considered. Data en-velopment analysis indicates that the conventional banks recovered better from the 2008 financial crisis, both in terms of cost and profit efficiency. Stochastic frontier analysis re-sults are mostly insignificant for our sample. Only one model shows significant rere-sults, indicating a better recovery for conventional banks regarding profit efficiency. To sum up, the results we find are mostly in favor of the conventional banks in the sample, ex-cept for the performance on asset-quality. Therefore, we conclude that the conventional banks in our sample recovered better from the 2008 financial crisis.

Keywords: banking, cost efficiency, profit efficiency, performance, Islamic banks,

iii

Acknowledgements

First of all, I would like to thank dr. J.C.M. van Ophem, my thesis supervisor of the University of Amsterdam for his time and contribution. During the sessions we had, he challenged my research constantly in such a way that it sharpened my critical thinking. This really motivated me to make my research more complete. Furthermore, I would like to thank dr. S.A. Broda, the second reader of my thesis of the University of Amsterdam, for attending my thesis presentation and taking the time to read my thesis.

I would also like to thank Ted van der Aalst PhD and Peter Bosschaart MSc, my the-sis mentors from KPMG, for their constructive feedback and for thinking along with the problems I faced during the writing process. Their time and comments greatly helped in raising the level of this thesis. Furthermore, I would like to thank the rest of the col-leagues from KPMG FRM for their interest in my thesis topic and the good atmosphere within and outside the office.

I must express my gratitude to René Pot and Roel van Pel for providing me with support during our regular Skype sessions. Furthermore, I would like to thank my fel-low students and roommates for the continuous encouragement throughout my years of study and for their company outside the study. This accomplishment would not have been possible without them. Thank you.

iv

Contents

2 Summary and literature review 5 2.1 Islamic banking summary . . . 5

2.2 Literature review . . . 8

3 Data and methodology 13 3.1 Banks included in the analysis . . . 13

3.2 Variables . . . 14

3.3 Descriptive statistics . . . 16

3.4 Methodology . . . 21

4 Empirical results 37 4.1 Distinguishability between Islamic and conventional banks . . . 37

4.2 Performance of the two banking systems . . . 41

4.3 Efficiency of the two banking systems . . . 45

5 Conclusion 56 5.1 Conclusion . . . 56

5.2 Limitations and further research . . . 57

Appendix . . . 59

v

List of abbreviations

CB conventional banks

COSR cost to income

CTA cash to assets

CTD cash to deposits

DA debt to assets

DEA data envelopment analysis

DFA distribution-free analysis

DTA deposits to assets

EM equity multiplier

EQL equity to net loans

EQTA equity to assets

FRA financial ratio analysis

GCC Gulf Cooperation Council (Bahrain, Kuwait, Oman, Qatar, Saudi Arabia, and the United Arab Emirates)

IB Islamic banks

IMF International Monetary Fund

IMLGL total impaired loans to gross loan ratio

LdASF liquid assets to deposits short-term funding ratio

LDBR net loans deposits borrowing

LR loan ratio

LTD loans to deposits

NetLTA net loans to assets ratio

PEA provision to earning assets

vi

ROA return on assets

ROD return on deposits

ROE return on equity

SFA stochastic frontier analysis

TFA thick frontier analysis

TLE liabilities to equity

WRL write-off ratio on loans

List of definitions

Iljara Leasing or renting; providing services and goods temporarily for a wage

Istisna A commission to purchase raw materials for manufacturing, paid

periodically

Mudaraba A partnership where capital is provided, in cash or assets (no debt is

accepted) by one party and labor is provided by the other party

Murabaha A contract of sale between the bank and its client for the sale of goods at a

price plus an agreed profit margin for the bank

Musharaka A classical partnership agreement. All parties involved contribute towards

the financing of a venture

Salam A commission to purchase raw materials for manufacturing, fully paid

upfront

Wakala Used in Islamic finance to describe a contract of agency or delegated

authority pursuant to which the principal appoints an agent to carry out a specific task on its behalf

1

Chapter 1

Introduction

1.1

Problem setting

The financial crisis, starting in 2008, influenced financial institutions all over the world. Commercial banks, investment banks, insurance companies, and brokerages were all hit by the crisis. Different types of institutions were hit in different ways and it is interesting to examine how these organizations reacted to the crisis. In this thesis, we focus on two different types of financial organizations: Islamic banks and conventional banks. The way these two banking systems operate differs from each other: the rules under which they operate, the revenue model and the relation between bank and client is different. Therefore, it is interesting to compare how these two types of banks react to a financial crisis.

Islamic banks ask no interest for loans and no interest is given on saving accounts. Instead, their revenue model is based on PLS, profit and loss sharing. Furthermore, Is-lamic banks are not allowed to invest in businesses that are not halal. Halal can be trans-lated as lawful. Hence, the rules under which they operate are more restrictive. Given these differences, it is worthwhile comparing Islamic banks with conventional banks. Should financial crises have a lower impact on the performance of Islamic banks and should they recover faster from a crisis, it might be interesting for conventional banks to adapt features of Islamic banks by adopting one of the three types of Islamic financial institutions:

• full-fledged Islamic financial institutions;

• Islamic windows in conventional financial institutions; • Islamic subsidiaries of conventional financial institutions.

Several studies investigate conventional banks and Islamic banks before and during the financial crisis of 2008. Iqbal (2001) examines Islamic banking in the nineties and con-cludes that Islamic banks are well capitalized, profitable and stable in general. Besides that, profitability ratios are more favorable compared to conventional banks, where the latter bear less risk, since the risk management instruments that can be used are more extensive compared to Islamic banks. Hasan and Dridi (2010) investigate how Islamic

Chapter 1. Introduction 2

banks and conventional banks performed during the 2008 global crisis by looking at prof-itability, credit growth and asset growth in several countries where the banking systems have a significant market share. They conclude that factors related to the Islamic banking business model helped in limiting the adverse impact on profitability in 2008, but that there has been a larger decline in profitability in 2009 compared to conventional banks due to weaknesses in risk management in some Islamic banks. Islamic bankings’ credit growth and asset growth performed better compared to conventional banks in both 2008 and 2009, contributing to financial and economic stability. Khediri, Charfeddine, and Youssef (2015) investigate the features of Islamic and conventional banking in Gulf Co-operation Council (GCC)1 countries over the period 2003–2010, hence pre-crisis, during crisis and a small period post-crisis. They conclude that the profitability, liquidity and capitalization for Islamic banks is better on average and that the credit risk is lower com-pared to the credit risk of conventional banks. Furthermore, they find that Islamic banks are less involved in off-balance sheet activities and have more operating leverage than their conventional counterparts on average. On the other hand, Ariss (2010) investigates 250 Islamic and conventional banks between 2000 and 2006 and found no direct evidence showing that Islamic banks are more profitable than conventional banks.

There are several demographic regions that are interesting to investigate, either be-cause Islamic banking is well represented in the banking sector, or there is a large growth potential, or banks can tap into new markets. The International Monetary Fund (IMF), states that Islamic banking becomes systematically important in Asia and the Middle East.2 Furthermore, they conclude that some products, such as the Islamic equivalent of bonds (Sukuk), expand significantly across the world. There is evidence that in GCC countries Islamic banking exceeds one third of total market share.3 _{Furthermore, there}

are several countries in Asia and the Middle East with a large Muslim population where the Islamic banking system is emerging or untapped.4 _{In the United Kingdom, Islamic}

banks account for 1% of the assets in the banking sector. In the rest of Europe, there are several developed financial markets, with a wealthy Muslim minority, where Islamic banking is not yet possible, since there are no banks that provide Islamic banking.5

Given the different demographic regions, the GCC region is worth studying because of the following reasons: I) the market share of Islamic banking is relatively high in the GGC region compared to the market share of Islamic banks in other countries and II) the banking system is relatively developed in comparison with the banking systems of other countries in the Middle-East.

1_{Alphabetically: Bahrain, Kuwait, Oman, Qatar, Saudi Arabia and United Arab Emirates} 2_{www.imf.org/external/themes/islamicfinance/}

3_{According to the World Islamic Banking Competitiveness Report 2016 from EY} 4_{E.g. China, India, Indonesia and Turkey}

Chapter 1. Introduction 3

1.2

Research question

Although the literature pre-crisis and during the crisis on Islamic banking is extensive (see Section 2.2), less is known on the period post-crisis. Among others, Khediri, Charfed-dine, and Youssef (2015) incorporate a post-crisis period in their analysis, however, only the years 2009 and 2010 are included. Therefore, to fill the gap in the literature, this the-sis investigates the post-crithe-sis period. Furthermore, the pre-crithe-sis and crithe-sis periods are included to obtain an overview of the development of the Islamic and the conventional banking systems over time. Given the previous stated problem and setting, the following research question is constructed: Did the Islamic banking system or the conventional banking system recover better from the 2008 financial crisis? For this analysis, we concentrate on the GCC countries from 2006 to 2015, where both banking systems have significant market share.

To answer the research question, three subquestions are defined:

• Question 1: Can we distinguish between Islamic and conventional banks based on financial ratios?

The reason for this research question is that we want to know whether financial ratios are good discriminators between both banking systems in the GCC region. A logit model is used to investigate whether we can differentiate between the two types of banks based on financial ratios. Among others, Olson and Zoubi (2008) include this analysis in their work before comparing the two banking system. Fur-thermore, they state that the logit model can be used out-of-sample to distinguish between Islamic and conventional banks based on financial ratios. Using a within sample and an out-of-sample test, we examine if we obtain the same results. • Question 2: How did the conventional and Islamic banks in the sample recover from the

2008 financial crisis in terms of profitability, liquidity, risk and asset-quality?

Next, we investigate the performance of Islamic and conventional banks in terms of financial performance. In the literature, financial ratio analysis (FRA) and logit models (refer to Question 1) are used mostly to compare the two banking systems. Among others, Khediri, Charfeddine, and Youssef (2015) use both techniques for their analysis. Similar to Khediri, Charfeddine, and Youssef (2015), we use a uni-variate analysis (FRA) and a logit model to compare the performance of the two banking systems.

Chapter 1. Introduction 4

• Question 3: Which banking system operates more efficient within GCC, the Islamic or the conventional banking system?

Efficiency is important in the banking sector, given the fact that resources are lim-ited and one wants to obtain the best results given these limlim-ited resources. In this thesis we focus on cost and profit efficiency. To answer this question, nonparametric data envelopment analysis (DEA) and parametric stochastic frontier analysis (SFA) are used and compared. Farrell (1957) and Charnes, Cooper, and Rhodes (1978) are considered the founders of DEA, a technique that is widely used to investigate the efficiency of banking systems and banking branches. Aigner, Lovell, and Schmidt (1977) and Meeusen and Broeck (1977) introduced SFA, which is the parametric counterpart of DEA.

Combining the outcomes of the three subquestions allows us to answer the research question of this thesis: Did the Islamic banking system or the conventional banking sys-tem recover better from the 2008 financial crisis?.

Based on the previous literature, that is discussed in the next chapter, the hypothesis is that it is possible to use financial ratios to distinguish between conventional and Islamic banks both within sample and out-of-sample. Furthermore, the hypothesis is that Islamic banks are less efficient compared to their conventional counterparts, however, we think that they recovered faster from the 2008 financial crisis due to a relative increase in profitability, liquidity and asset-quality. Regarding risk, we think that conventional banks recovered better.

In several papers, authors make the same comparison pre-crisis and during crisis. This thesis contributes to the most recent literature by investigating the recovery of the fifteen largest Islamic and fifteen largest conventional banks in the GCC region after the 2008 financial crisis. We choose to include the largest banks from both banking systems since we believe that these banks are each other competitors.

Reading guide

The remainder of this thesis is constructed as follows. Chapter 2 contains a summary of the concept of Islamic banking and a literature review. In Chapter 3, the description of the data and methodology that is used can be found. Chapter 4 contains the results, followed by the conclusion in Chapter 5.

5

Chapter 2

Summary and literature review

2.1

Islamic banking summary

2.1.1 The founding of Islamic finance

The precise origin of Islamic banking is uncertain. Wilson (1983) states that the first interest-free bank was constituted in rural Pakistan in the late 1950s. Di Mauro et al. (2013), a paper from the European Central Bank, examines the development of Islamic fi-nance with a focus on Europe. They state that the first Islamic banks started in Egypt and Malaysia in 1963. According to Khan (2013), Islamic banking spread around the world from 1975 onwards. In the next 20 years, the sector grew tremendously, up to 144 Islamic financial institutions in 1995 according to Kepel (2006). According to ’The Banker’1_{, in}

2015, Islamic Financial Institutions are represented in 37 countries, with 198 commercial banks and 360 institutions reporting a total of US$ 1,273 billion assets. Deutsche Bank estimates the size of the global banking sector in 2014 at US$ 294 trillion.2 Estimations on the total number of banks worldwide are absent, however, there are sources that estimate the number of banks in Europe around 9,000 and in the United States around 7,000.3

2.1.2 Prohibited and permitted financing methods

According to El-Gamal (2000), there are two fundamental prohibitions in Islamic finance compared to conventional finance. First of all, there is the prohibition of ’riba’, which can be translated as usury or interest. Hence, it is not allowed to charge extra money for any outstanding debt. Additionally, it was the case that it was not allowed to reduce debt by prepayment. However, in recent years, regulation around reducing debt using prepayment has been relaxed. Another debt-related topic that caused discussion is the re-sale of debt. In Malaysia, this is a widely used concept, which is strictly forbidden in other countries.

Second, there is the prohibition of ’gharar’, which can be translated as risk or uncer-tainty. The prohibition of gharar implies that one cannot sell a good that does not exist,

1_{Top 500 Islamic financial institutions,}

http://www.thebanker.com/Banker-Data/Top-Islamic-Financial-Institutions

2_{http://www.businessinsider.com}

Chapter 2. Summary and literature review 6

does not exist yet or is not in the possession of the seller. To illustrate: a fish can only be sold when it is already in the store, not when it is still in the sea. The prohibition of gharar has far-reaching consequences, and forbids a number of financial instruments that are common in conventional banking. Two of these fields of financial instruments are in-surance and financial derivatives. Inin-surance is not allowed since an insured person pays monthly premiums without the assurance that this money will be repaid to him/her. Furthermore, there is the problem that insurers cannot obtain interest on the premium that clients pay to them. This is a problem since the business model of insurers is partly based on receiving interest on the premiums paid. Especially the business model for life-insurers, where the time between the premium received and payment to the client is relatively long, is based on a positive interest rate. Next to insurance, financial derivatives are prohibited: forwards, futures, options and other derivatives. Forwards and futures are not allowed since the product of sale is non existing at the time that the trade is made. Options are not allowed since there is risk involved.

This leaves the discussion whether or not investing in stocks is allowed in Islamic finance. The answer is that it is allowed to invest in stocks, given that certain conditions are met and the potential profit is halal. A profit or business is halal whenever it is compliant with the sharia law. The reasoning behind the allowance of stocks is that, whenever you obtain stocks, you are a partner in the business and you share in the losses and profit. This equates to the Islamic concept of ’musharaka’, which is discussed next, and is halal. The conditions that have to be met are, among others, that the company that offers the stocks has to be halal and only cooperates with other halal firms. Furthermore, the concerning company must have a certain capital structure: the debt to asset ratio must be lower than 33%; non-operating interest income must be lower than 10% of total income and accounts receivable to total assets must be lower than 47%.4

In Islamic finance, most financing types are permitted, as long as there is no interest and uncertainty involved. To make it understandable what is allowed and what not, several contracts are predetermined.

• Cost-plus sales and credit sales (murabaha)

In the case of cost-plus sales and credit sales, the seller obtains the product for a certain price and the buyer and the seller agree that the buyer pays the initial price plus a premium. This premium can be a lump-sum or a percentage. In Islamic finance, there is a difference between the following two cases: (I) agreeing to pay $11 for a product now that has a value of $10, and (II) a credit sale for $10 that will be postponed with a price raise to $11. Although some argue that both trades can be seen as a payment with interest, the general consensus on this issue is that murabaha is an increase of price, which is allowed, and interest is an increase of

4_{The capital structure is obtained from the institute of Islamic banking and insurance}

Chapter 2. Summary and literature review 7

debt, which is not allowed. This implies that (I) is allowed in Islamic finance, while (II) is not allowed.

• Leasing

Leasing is the sale of the right to use the object for a defined period of time instead of the sale of the object itself. There is a difference between conventional leasing and Islamic leasing. In Islamic finance, the leasing company must own the product that it leases to its client, while in conventional finance, a leasing company is allowed to lease products to clients that are owned by a third party. In the case of mortgages, the Islamic bank buys the house and is the owner. The client makes payments and when the price of the house plus a fixed markup is repaid, the buyer gains complete ownership of the house. The other possibility is that the client leaves the house and gives the property back to the bank.

• Partnerships (mudaraba and musharaka)

There are two types of partnerships in Islamic finance, ’mudaraba’ and ’musharaka’. The first is a silent partnership where one partner is the investor and one partner is the working partner. The latter is a full partnership. These partnerships are based on the PLS, profit and loss sharing, principle. Next to the general musharaka, there is the ’mushakara mutanaqisah’, which can be translated as diminishing partner-ship. In this form, the ownership is shared between the lessor and the lessee. Cus-tomer payments are partly a rental payment and partly a buy-out of a part of the ownership. This continues until the customer completely paid the asset price and is the owner of the product.

• Islamic forwards

There are two types of Islamic forwards that are legally allowed, ’salam’ and ’is-tisna’. Salam is a commission to manufacture. The full price is paid initially such that the manufacturer can purchase raw materials (seeds) to produce products (fruit). Istisna provides financial resources for production, however, the payments are in installments as the work progresses. There are many conditions that have to be fulfilled before a salam or istisna contract is legally approved.

• Sukuk

Sukuk are the Islamic equivalents to bonds and the sukuk market has grown tremen-dously since 2001. Sukuk (plural and singular) is defined by the Accounting and Auditing Organization for Islamic Financial Institutions (AAOIFI) as:

‘certificates of equal value representing undivided shares in the ownership of tangible assets, usufructs and services or (in the ownership of) the assets of particular projects or special in-vestment activity’.

Chapter 2. Summary and literature review 8

According to the Autoriti Monetari Brunei Darussalam5, the sukuk market grew from US$ 14.8 billion in 2001 to US$ 281.3 billion in 2013. As of 2015, in different countries, such as United Kingdom, Luxembourg, Ireland and Hong-Kong, where the Muslim community is a minority sukuk are issued.

2.2

Literature review

As long as Islamic banking exists, papers have been written comparing the performance of Islamic banks and conventional banks. In this section, several topics regarding Islamic finance are discussed. First of all, we discuss literature regarding the business models of Islamic banks and conventional banks. Furthermore, we discuss the literature that deals with distinguishing between both banking systems based on financial ratios. Next, we review the performance of Islamic banks pre-crisis and during the crisis. Finally, provide a literature review on efficiency in banking.

2.2.1 Comparing Islamic and conventional banks in the GCC region

To be able to directly compare the Islamic and conventional banks in this thesis, we have to ensure that the two banking systems are related. The banks chosen for the analysis, refer to Section 3.1, are the fifteen largest banks from both systems in the GCC region. One may argue that direct comparison of conventional and Islamic banks is not possible since the business models of the banks are too different, or that the environment in which they operate is not the same. The difference in business models is that the motivation for conventional banks is to make profit and that the goal of Islamic banks is more towards social value and ethical behavior.

We propose that comparison between the two banking systems is justified. The first argument comes from Beck, Demirgüç-Kunt, and Merrouche (2013). They state that in theory, the two banking systems are significantly different. These differences are dis-cussed in Section 2.1.2. However, in reality, some financial products used in Islamic finance mirror financial products in the conventional banking system. Discounting is replaced with contingent payment structures and commission on price replaces the in-terest rate payments. Furthermore, financial products related to leasing are widely used among Islamic banks, as they are directly comparable to conventional banking contracts. This does not imply that there are no differences. There is still the risk of profit and loss sharing that Islamic banks and clients take that has implications for the financial position of the Islamic banks.

The second and the third argument are given by Johnes, Izzeldin, and Pappas (2014). Earlier, the two banking systems operated in different markets, which implies that a client could only choose one type of bank, a conventional or an Islamic bank. However, these

Chapter 2. Summary and literature review 9

days, the competition between the two banking systems is increasing. In the five coun-tries considered, both banking systems have considerable amount of market share.6 The national market shares of the Islamic banks, based on banking assets are given by: Saudi Arabia (51.2%), Kuwait (45.2%), Bahrain (29.3%), Qatar (25.8%) and United Arabic Emi-rates (21.6%). Johnes, Izzeldin, and Pappas (2014) conclude that direct competition be-tween Islamic and conventional banks in the same markets allows a direct comparison between the two systems. The reason is that since the banking systems can coexist and consumers choose the bank of their preference, there is a basis for comparison. As a sec-ond reason they state that: ’DEA, by estimating a frontier which envelops the observed production points with piece wise linear segments, allows each bank to have its own objectives as it is only compared with banks with a similar input and output mix’.

The last argument that the banking systems can be compared directly is that the banks have a lot in common. The banks all operate in the same region, are subject to the same international accounting standards, have a considerable amount of market share since they are the largest in their countries based on the amount of assets and are multi billion dollar companies.

One could argue that, since the banks operate in the same region, the clientele of the banks could be biased. Since the majority of the population in GCC countries is Muslim, the hypothesis could be that all Muslims want to be the client of an Islamic bank and that the Islamic banks can choose the most profitable clients in their portfolio. Examining the literature on this matter, no research was found that discusses this problem. The problem regarding the client portfolio of banks is discussed from another angle in the literature. Among others, Gerrard and Barton Cunningham (2001) and Ahmad and Haron (2002) discuss why clients choose an Islamic or a conventional bank and not how Islamic and conventional banks accept or reject clients. This could mean two things: either, it is not the case that Islamic banks only accept the most profitable clients, or, this problem is neglected in the literature.

2.2.2 Distinguishing between Islamic and conventional banks

There are several methods to examine whether one can distinguish between Islamic and conventional banks based on financial ratios. This method is based on the test for struc-tural differences between the financial ratios of both banking types. Metwally (1997) uses three different methods for their analysis. Next to a probit model and discriminant analy-sis, a logit model is performed. They conclude that liquidity, leverage and credit risk can be used to discriminate between the two type of banks, however, that profitability and ef-ficiency cannot be used. Khediri, Charfeddine, and Youssef (2015) use linear discriminant analysis and a logit model to distinguish between the two banks and Olson and Zoubi (2008) use a logit model, neural network, and k-means nearest neighbor classification model to assign a bank correctly to the conventional or the Islamic banking system. Since

Chapter 2. Summary and literature review 10

Metwally (1997), Olson and Zoubi (2008) and Khediri, Charfeddine, and Youssef (2015) all use a logit model to distinguish between the both banking systems and we have the data available for the logit model, we choose the logit model to check whether we can distinguish between Islamic and conventional banks.

2.2.3 Performance of Islamic banks, pre- and during crisis

There are several papers written on the performance of Islamic banks, however no con-sensus has been reached on which banking system performs better. Since we focus on the categories profitability, liquidity, risk and asset-quality, we focus on the literature concern-ing these financial ratios. Iqbal (2001) concludes that Islamic banks are well capitalized, profitable and stable compared to conventional banks, however, conventional banks are more cost effective. He concludes this based on financial numbers and ratios of twelve banks in the Middle East from 1990 to 1998. Samad (2004) examines the comparative performance of Bahrain’s interest-free Islamic banks and the interest-based conventional commercial banks during the post Gulf War period with respect to three aspects: prof-itability, liquidity risk, and credit risk. He advocates for using ratio analysis to measure bank performance, since ratios remove disparities, such as size, between banks. Further-more, he concludes that Islamic banks are superior regarding credit risk, however, there is no significant difference in profitability and liquidity risk. To measure profitability, they use ROA, ROE and COSR. To address liquidity, they investigate LT A, LdASF and LDBR. Finally, for credit risk, EQT A, EQL and IM LGL are used (refer to the list of abbreviations).

Hasan and Dridi (2010) examine the impact of the 2008 financial crisis on profitabil-ity, credit growth, asset growth and external ratings. For the analysis, 120 Islamic and conventional banks from eight different countries are used, where both Islamic banks and conventional banks are well represented. These countries are Jordan, Malaysia, Turkey and the GCC countries, excluding Oman. They conclude that the business models of Islamic banks have an ambiguous effect on the profitability. In 2008, Islamic banks per-formed better in terms of profitability. The reason that is given is that Islamic banks have no exposure to toxic assets, derivatives, and conventional securities. However, the prof-itability of conventional banks outperformed the profprof-itability of Islamic banks in 2009, since Islamic banks were unable to manage the risk of the crisis. In terms of credit per-formance, Islamic banks outperformed conventional banks in all countries between 2007 and 2009. For asset growth, the same result yields. Hasan and Dridi (2010) highlight that Islamic banks face several challenges, exposed by the crisis:

• Tools for liquidity risk are underdeveloped, since several tools conventional banks use are prohibited by sharia law;

• Regulation is insufficient;

Chapter 2. Summary and literature review 11

• The knowledge and expertise did not grow in the same pace as the sector.

Khediri, Charfeddine, and Youssef (2015) investigate Islamic and conventional banks in GCC countries over the period 2003 - 2010. Hence, both pre- and during crisis. They test three hypotheses: I) Islamic banks are more profitable than conventional banks, II) Islamic banks hold a higher amount of liquid assets compared to conventional banks, III) Islamic banks are less risky compared to conventional banks. These hypotheses are tested using ratios categorized in five segments: profitability, liquidity, credit risk, in-solvency risk and asset structure. The ratios are comparable with the ratios chosen by Samad (2004). For the analysis, they use parametric methods (linear discriminant analy-sis and logistic regressions) and nonparametric methods (neural network method and a classification tree). Khediri, Charfeddine, and Youssef (2015) find that there is a difference in profitability, liquidity, credit risk and capitalization, where Islamic banks outperform conventional banks in the GCC region. Furthermore, they stress that there is room for improvement in the supervisory framework of Islamic banks.

Abedifar, Molyneux, and Tarazi (2013) define three econometric models, one for credit risk, one for insolvency risk and one for bank interest rates to compare Islamic and conventional banks in 24 countries in the Middle East from 1999 to 2009. These models are used to determine loan quality, stability and to capture any special rents extracted by Islamic banks from their clients for their Sharia compliant services. Dependent variables are, among others: loan loss reserves to gross loans, the Zscore, which is a ratio involving the expected ROA and the standard deviation of the ROA, and net interest margins. Furthermore, dummy variables such as country and year are used. Lastly, there is a set of control variables such as size, market share, and capital asset ratio. The outcome of their research is that Islamic banks have lower credit risk, especially when their size is small and/or the country they operate in has a Muslim population exceeding 90%. Furthermore, they find that Islamic banks are more stable in terms of insolvency risk.

2.2.4 Efficiency in banking

Since the financial sector is a highly competitive sector, the players in the sector need to be efficient. With efficiency we mean a process that uses the lowest amount of inputs to create the greatest amount of outputs. In the literature, a number of efficiency types are present. The most common types of efficiency that are discussed are technical, alloca-tive efficiency, cost efficiency and profit efficiency. In their book, Kumbhakar and Lovell (2003) discuss the different types of efficiency. Technical efficiency measures whether the amount of output of a company given the amount of input is the maximum feasible output. If this is the case, than the technical efficiency equals one and otherwise it is a number between zero and one. Allocative efficiency measures how good the allocation of inputs is given prices and production technologies. For comparing efficiency within and between banks, cost and profit efficiency are the most relevant efficiencies to investigate. These metrics measure how efficient banks are, and with efficiency we mean how low the

Chapter 2. Summary and literature review 12

costs or how high the profit of the bank is, given the selected input and output variables. Ferrier and Lovell (1990), Ferrier et al. (1993), Cummins and Zi (1997) Kuosmanen and Post (2001) choose the cost based approach. The profit based approach is discussed in Cooper, Seiford, and Zhu (2004) and Coelli et al. (2005). Srairi (2010) uses both cost and profit efficiency of conventional and Islamic banks in GCC countries for his paper.

To calculate the efficiency of banks, there are two main techniques used in the litera-ture: nonparametric data envelopment analysis and parametric stochastic frontier analy-sis. Both techniques rely on input and output variables. To provide an idea of the variety of input and output variables used in the literature on banking efficiency refer to Table 2.1. Note that these days, most of the input and output variables are financial numbers where earlier variables such as computer terminals or number of offices where used.

TABLE 2.1: Review of used input and output variables in the bank effi-ciency literature

Paper Input variables Output variables Vassiloglou and Giokas (1990) Labor, supplies, ranch installation, computer terminals Number of transactions Seiford and Zhu (1999) Employees, assets, stockholders’ equity Revenue, profit

Yudistira (2003) Staff costs, fixed assets, total deposits Total loans, other income, liquid assets Luo (2003) Employees, assets, equity Revenue, profit

Sathye (2003) Model 1: Interest expenses, non-interest expenses Model 2: Deposits , staff numbers

Model 1: Net interest income, Non-interest income Model 2: Net loans, Non-interest income Wu, Yang, and Liang (2006) Personnel, other general expenses Deposits, loans, revenue

Mokhtar, Abdullah, and Al-Habshi (2006) Total deposits, total overhead expenses Total earning assets

Avkiran and Fukuyama (2008) Interest expenses, non-interest Expenses Interest income, non-interest income Sufian (2008) Total deposits, assets Total loans, income, investments Morita and Avkiran (2009) Interest expenses, non-interest expenses Interest income, non-interest income Staub, Souza, and Tabak (2010) Operational expenses, personnel expenses Deposits, loans, revenues

13

Chapter 3

Data and methodology

3.1

Banks included in the analysis

To compare the two banking systems, the fifteen largest Islamic banks and conventional banks in the GCC region (based on the banks assets in the year 20151) are included in the analysis. For each country, the number of banks in the analysis is based on the number of banks that are in the top 15 of the Islamic banking system. For example, since there are two banks from Bahrain in the 15 largest Islamic banks, the two largest conventional banks in Bahrain are included in the analysis. Alternatively, we could choose to include the fifteen largest banks of each system in the whole region, however, in that case, coun-try specific effects do not have a symmetric effect on the results. In case of an event in a country that reduces profitability or efficiency, the result could be biased towards the banking system that is the most prevailing. For example, if there are four conventional banks and two Islamic banks and such an event occurs, the result would suggest that Islamic banks in the data set performed better relative to the conventional banks. How-ever, the underlying reason is the country specific effect, and not the difference between the two systems. In Table 3.1 and Table 3.2, the banks included in the analysis are stated.

TABLE3.1: Largest fully fledged Sharia compliant banks in the GCC region

based on the total assets in 2015

Country Number of banks Banks included Total assets (2015)_{in Billion $}

Bahrain 2 Albaraka Banking Group 21.3

Al Salam Bank Bahrain 5.2

Kuwait 2 Kuwait Finance House 56.8

Boubyan Bank 8.8

Qatar 2 Qatar Islamic Bank 26.4

Qatar International Islamic Bank 10.6

Saudi Arabia 4 Al Rajhi Bank 82.0

Alinma Bank 21.6

Bank Aljazira 17.7

Bank Albilad 12.1

United Arabic 5 Dubai Islamic Bank 33.7

Emirates Abu Dhabi Islamic Bank 30.5

Emirates Islamic Bank 11.7

Noor Bank 7.9

Sharjah Islamic Bank 7.1

Mean 23.6

Chapter 3. Data and methodology 14

TABLE 3.2: Largest conventional banks in the GCC region based on the total assets in 2015

Country Number of banks Banks included Total assets (2015)_{in Billion $}

Bahrain 2 Ahli United Bank 33.4

Arab Banking Corporation 29.4

Kuwait 2 National Bank of Kuwait 74.3

Burgan Bank 26.4

Qatar 2 Qatar National Bank 133.6

Commercial Bank of Qatar 31.8

Saudi Arabia 4 National Commercial Bank 116.0

Samba Financial Group 58.0

Riyad Bank 57.2

Banque Saudi Fransi 50.3

United Arabic 5 National Bank of Abu Dhabi 102.4

Emirates Emirates NBD 98.8

First Gulf Bank 57.7

Abu Dhabi Commercial Bank 55.5

Mashreq Bank 28.8

Mean 63.6

Table 3.3, shows the number of Islamic banks and conventional banks in the sample.

TABLE3.3: Sample size Islamic and conventional banks

2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 Total Islamic 10 12 13 13 14 15 15 15 15 15 137 Conventional 14 13 15 15 15 15 15 15 15 15 147 Total 24 25 28 28 29 30 30 30 30 30 284

There are two reasons why there are missing values and the total sample does not add up to 300 banks. First, not every bank in the sample existed in 2006. For banks that where founded between 2006 and 2015, we excluded the first year after founding from the analysis since we noticed that the financial ratios of these starting years where highly different from the other years in the sample. On the other hand, not all annual reports or consolidated financial statements could be found. In total, we have 284 data points for the analysis, of which 137 Islamic banks and 147 conventional banks. From 2011 onwards, all annual reports are available.

3.2

Variables

For the analysis, data available from 2006 to 2015 on several economic indicators is used. These indicators demonstrate the financial position of conventional and Islamic banks. The data that is used differs for each research question. For subquestion 1 regarding distuingishability and subquestion 2 regarding performance, the indicators that are used are ratios, so we do not face the problem that banks in different countries use different currencies. These ratios are given in Table 3.4. For subquestion 3, the variables used for the analysis are given in Table 3.5.

A problem that could arise in the data is that the format of the annual reports of Islamic and conventional banks differ. Conventional banks in the GCC region comply to IAS (International Accounting Standards) regulation. On the other hand, Islamic banks

Chapter 3. Data and methodology 15

comply to accounting rules following the Islam religion from the AAOIFI (Auditing Or-ganization for Islamic Financial Institutions). Furthermore, every Islamic bank has a Shariah Supervisory Committee to guarantee that every transaction is compliant with Islamic laws. AAOIFI and IAS standards are not the same. However, all banks in the GCC region must comply with the regulation of the central banks of the country they op-erate in. Given that in each GCC country, the central banks have adopted IAS regulation, all consolidated financial statements are subject to IAS regulation and hence the financial numbers are comparable between banks, between banking systems and between coun-tries.

Considering Table 3.4, the third column indicates whether an increase in the ratio has a positive or a negative effect on the performance of the category. Banks want all variables regarding profitability and liquidity to be as high as possible and variables re-garding risk and asset-quality to be as low as possible, with an exception for the variable CAR, which defines the capital adequacy ratio, a measure used in Basel III, which is the global regulatory framework on bank capital adequacy, liquidity risk and stress testing.

TABLE3.4: Variable definition for subquestion 1 and subquestion 2

Category Variable Effect Name Formula

Profitability ROA ↑ Return on assets net income/total assets ROE ↑ Return on equity net income/total equity PM ↑ Profit margin net income/operating income ROD ↑ Return on deposits net income/total deposits Liquidity CTA ↑ Cash to assets cash/total assets

CTD ↑ Cash to deposits cash/total deposits Risk DTA ↓ Deposits to assets total deposits/total assets

CAR ↑ Capital adequacy ratio Tier I & II capital / risk weighted assets DA ↓ Debt to assets total liabilities/total assets

EM ↓ Equity multiplier total assets/stockholders equity TLE ↓ Total liabilities to equity total liabilities/stockholders equity Asset-quality LR ↓ Loan ratio loans and advances/total assets

LTD ↓ Loans to deposits loans and advances/ total deposits

To construct the ratios, the following variables are obtained: net income, total assets, total equity, operating income, total deposits, cash, Tier I and Tier II capital, risk weighted assets, total liabilities, stockholder equity, total liabilities and loans and advances. The information that is needed to construct these ratios is conducted from consolidated financial statements of the banks. The consolidated statements of the financial position, which includes the data used for the research, are audited and signed by accountancy firms to ensure the quality of the data. Since Islamic financial institutions do not use loans in the same way as conventional institutions, according to Olson and Zoubi (2008), the Islamic counterpart of loans and advances is equivalent to Murabaha, Mudaraba and Mushakara investments.

Considering Table 3.5, the left columns are the variables for the DEA and the right columns are the variables for the SFA. For these variables, a bank wants the output vari-ables as high as possible given the input varivari-ables. The input and output varivari-ables are

Chapter 3. Data and methodology 16

based on Dietsch and Lozano-Vivas (2000), Maudos et al. (2002), Yudistira (2003), Semih Yildirim and Philippatos (2007) and Srairi (2010). Besides the dependent variables Cs_and

Ps, one can see that there are 12 (4) variables used in the data envelopment (stochastic frontier) analysis. The reason that there are less variables used in the SFA is twofold. First of all, only output quantities (y) and input prices are used (p) in the literature. Further-more, in the SFA, second order and cross terms are used, which implies that the model becomes too large when there are more variables included. Since the number of banks is limited in this thesis, incorporating more variables in the SFA model could give the problem of overfitting.

TABLE3.5: Variable definition for subquestion 3

DEA Variable SFA Variable

Input xd

1 Fixed assets Efficiency Cs Cost

xd

2 Personnel expenses Ps Profit

xd₃ Deposits Output y₁s Loans Price input pd

1 Price of capital y2s Other earning assets

pd

2 Price of labor Price input ps1 Price of capital

pd₃ Price of deposits ps₂ Price of labor Output yd

1 Loans

yd₂ Other earning assets yd₃ Equity

Price output qd

1 Price of loans

q₂d Price of other earning assets qd

3 Price of equity

The price variables piand qiwith i = {1, 2, 3} are constructed as follows:

p1=

Other operating expenses Fixed assets p2= Personell expenses

Assets p3=

Total interest expenses Deposits q1=

Interest income Loans q2=

Other operating income Other earning assets q3=

Assets Equity

3.3

Descriptive statistics

3.3.1 Distinguishability and performance

For the financial ratios, which are derived from the financial numbers, the means per year are given in the Appendix A. For clarity, every category (prof itability, liquidity,

Chapter 3. Data and methodology 17

riskand asset-quality), has its own table. In Figure 3.1 up to Figure 3.4, the correspond-ing plots are given. The most strikcorrespond-ing are the decreascorrespond-ing ROA, ROE and ROD values for Islamic banks from 2006 to 2009 and the outliers regarding CT A and CT D for con-ventional banks in 2007. Examining the data, the first point can be explained by both large increases in assets and deposits and a small decrease in profit. The large increase in assets and deposits can be explained by an increase in popularity of the Islamic banking system. The asset and deposit levels of conventional banks where already on a higher level in 2006, and hence the drop in ROA, ROE and ROD is less severe. The outlier regarding CT A and CT D can be explained by two banks in the sample receiving more cash than usual compared to the other years.

FIGURE3.1: Mean financial ratios per year: Profitability

Chapter 3. Data and methodology 18

FIGURE3.3: Mean financial ratios per year: Risk

Chapter 3. Data and methodology 19

3.3.2 Efficiency

The descriptive statistics for the input and output variables for the third subquestion (Which banking system operates more efficient within GCC, the Islamic or the conven-tional banking system?) are monetary values and to compare them, all financial numbers are converted to US $ using exchange rates and consumer price index per December 31 of the year concerned.2 The historical exchange rates are given in Table 3.6. Note that only the Kuwait dinar is not directly linked to the US $ and fluctuates over time. The rest of the currencies are linked to the US $ with small fluctuations over time. The consumer price index per year and per country are given in Table 3.7. The year 2010 is used as reference year.

TABLE 3.6: Historical exchange rates Bahrain dinar (BD), Kuwait dinar (KD), Qatar riyal (QAR), Saudi Arabian riyal (SAR) and Arab Emirates

dirham (AED)

BD KD QAR SAR AED

2006 2.656 3.459 0.275 0.267 0.272 2007 2.663 3.658 0.275 0.268 0.272 2008 2.652 3.620 0.275 0.267 0.272 2009 2.652 3.484 0.275 0.267 0.272 2010 2.652 3.555 0.275 0.267 0.272 2011 2.653 3.593 0.275 0.267 0.272 2012 2.652 3.554 0.275 0.267 0.272 2013 2.652 3.543 0.275 0.267 0.272 2014 2.652 3.415 0.275 0.266 0.272 2015 2.652 3.294 0.275 0.266 0.272

TABLE3.7: Consumer price index (CPI) per year per country

Bahrain Kuwait Qatar Saudi United Arabic Arabia Emirates 2006 89.3 78.4 82.3 78.9 78.3 2007 92.2 82.7 93.6 82.2 87.0 2008 95.4 91.5 107.7 90.4 97.6 2009 98.1 95.7 102.5 94.9 99.1 2010 100.0 100.0 100.0 100.0 100.0 2011 99.6 104.9 101.9 105.8 100.9 2012 102.4 108.3 103.8 108.9 101.6 2013 105.8 111.2 107.1 112.7 102.7 2014 108.6 114.4 110.4 115.7 105.1 2015 110.6 118.2 112.5 118.2 109.4

2_{Exchange rates from http://www.xe.com/ and inflation consumer price index from}

Chapter 3. Data and methodology 20

The descriptive statistics of the used input and output variables are given in Table 3.8 for both the data envelopment and stochastic frontier analysis. Variables with a su-perscript d are used in DEA and variables with a susu-perscript s are used in SFA.

TABLE3.8: Descriptive statistics input and output variables data envelop-ment analysis and stochastic frontier analysis

Mean Standard deviation Minimum Maximum

IB CB IB CB IB CB IB CB Cs_{= Cost 582.11 1,195.50 530.48 531.10 65.61 367.22 2,269.92 2,776.31} Ps_{= Profit 370.21 865.52 510.81 552.94 -121.07 -140.83 2,466.39 2,766.89} xd 1= Fixed assets 387.72 355.17 609.94 190.20 2.07 105.06 2,668.49 837.31 xd 2= Personnel expenses 183.46 310.53 151.88 170.74 13.79 78.13 600.62 800.10 xd 3= Deposits 13,372.95 30,983.93 12,349.18 18,736.24 922.99 5,742.85 58,963.52 96,519.41 pd,s1 = Price of capital 0.832 0.725 1.222 0.310 0.074 0.184 10.818 1.635 pd,s2 = Price of labor 0.0093 0.0062 0.0078 0.0053 0.0041 0.0030 0.0143 0.0111 pd 3= Price of deposits 0.021 0.029 0.017 0.024 0.001 0.003 0.074 0.131 yd,s1 = Loans 11,261.63 28,091.55 10,406.26 16,375.81 316.93 5,793.147 48,152.49 94,834.65

yd,s₂ = Other earning assets 4,775.29 13,952.86 4,114.79 7,514.84 393.38 3,798.03 17,218.68 40,311.43 yd

3= Equity 2,669.38 6,104.44 2,335.05 2,943.67 479.56 1,827.54 10,508.92 15,156.12

qd

1= Price of loans 0.066 0.056 0.024 0.018 0.033 0.032 0.128 0.143

qd

2= Price of other earning assets 0.058 0.044 0.042 0.019 -0.003 0.012 0.294 0.108

qd

3= Price of equity 7.03 7.84 2.25 1.75 2.31 4.77 14.21 15.18

Prices are in US$ and all other variables are in US$ x1000

From 3.8, we can see that, based on the personnel expenses, deposits, loans, other earning assets and equity, the conventional banks in the sample are larger compared to the Islamic banks. Only the amount of fixed assets is larger for Islamic banks. In this case, being larger comes with more costs, which does not come as a surprise and more profits, which does not always have to be true. The data shows that the price of capital, labor, loans and other earning assets are higher for Islamic banks and that the price of deposits and equity are higher for conventional banks.

Chapter 3. Data and methodology 21

3.4

Methodology

To answer the research questions, the financial numbers and ratios stated in Section 3.2 are analyzed and a comparison between Islamic banks and conventional banks is made. In order to connect with the previous literature, the years 2006 - 2009 are included in the analysis and to add to the existing literature, the years 2010 - 2015 are included. First, we discuss the model used to answer subquestion 1 (distinguishability), logit modeling. Next, to answer subquestion 2 (performance), we discuss financial ratio analysis (FRA). Finally, for subquestion 3 (efficiency), we elaborate on two frontier techniques, nonpara-metric data envelopment analysis and paranonpara-metric stochastic frontier analysis.

3.4.1 Logit model

In the literature, there are at least three papers present that discuss the first research question: Can we distinguish between Islamic and conventional banks based on financial ra-tios? Metwally (1997) uses a logistic regression and both Olson and Zoubi (2008) and Khediri, Charfeddine, and Youssef (2015) use a logit model to distinguish between the two banking systems. This model is used in the latter two papers to examine the perfor-mance of both banking systems in terms of profitability, liquidity, risk and asset-quality. The model is given as follows:

log  pi 1 − pi  = α + n X j=1 βjxji+ i

In this model, we have as the dependent variable the logarithm of two probabilities, where pi ∈ [0, 1) is the variable of interest displaying the probability that a bank is an

Islamic bank. We examine this for every bank in four different time periods, the overall period, the pre-crisis period, the crisis period and the post-crisis period. We define i ∈ {1, 2, .., n} as the number of banks in the model. This implies that we have n = 284 (overall period), n = 49 (pre-crisis period), n = 56 (crisis period) and n = 179 (post-crisis period). Furthermore, α is a constant and xji and βj are the financial ratios and

the corresponding parameters respectively. Finally, iis the error term. The independent

variables are selected using a backward stepwise regression technique following Neter et al. (1996) and Khediri, Charfeddine, and Youssef (2015). As a convergence criterion, the Akaike information criterion (AIC) is used.

Multicollinearity

Before we estimate the model, we check if there is multicollinearity in the financial ratios. We use and compare two different methods to address the problem of multicollinear-ity. In the first method, we calculate a matrix with Pearson correlations and eliminate

Chapter 3. Data and methodology 22

the variables that are strongly correlated (a critical value of ρ = 0.7 is used following Hinkle, Wiersma, and Jurs (2003)). In the second method, we calculate the Variance Infla-tion Factor following Miles and Shevlin (2000), O’brien (2007) and Greene (1993). After eliminating variables that cause multicollinearity, we estimate the logit model with the remaining variables. If both methods result in different logit models, we test whether one of the models goodness of fit is better compared to the other model. We test this using a likelihood ratio test, with the test statistic being χ2-distributed.

For the Variance Inflation Factor, we start from a linear model y = Xβ + , where X ∈ Rnx(k+1) is a matrix with the first column of ones and the other k columns being the values of the k independent variables. The variance-covariance matrix of regression coefficients is given by:

σ2(β) = σ_2(X0X)−1 (3.1)

Then, according to and using the notation of O’brien (2007), the variance of the ith re-gression coefficient is given by:

σ2(βi,a) = σ2(Xi0MaXi)−1with (3.2)

Ma= I − PXa = I − Xa(X

0

aXa)−1Xa0 (3.3)

Where a 6= i represents the other independent variables and both Maand PXa are

pro-jection matrices, hence, symmetric and idempotent. Writing xxiias the inverse of the ith diagonal element of (X_i0MaXi)−1and using (3.2) and (3.3), O’brien (2007) states:

xxii = (x0_ixi− x0iPXaxi)

−1

This can be rearranged using division and multiplication by x0_ixias

xxii=  x0_ixi  1 −x 0 iPXaxi x0_ixi −1

In the last equation 1−x

0 iPXaxi

x0

ixi is the proportion of variance in the ith variable associated

with the other independent variables a. When we use that R2_i = 1 − x0iPXaxi

x0

ixi andP x

2 i

is the sum of squares for the ith independent variable, the last equation can be written without matrix notation as:

xxii= 1 (1 − R2 i)P x2i (3.4) Applying (3.2) and (3.4): σ2(βi) = σ_2 (1 − R2_i)P x2 i (3.5)

Chapter 3. Data and methodology 23

To obtain an unbiased estimator for the variance of the coefficients, we need an unbiased estimator for σ2

, which is given by:

ˆ σ2_ = P i(Yi− ˆYi)2 n − k − 1 = (1 − R2 y) P i(Yi− ¯Y )2 n − k − 1 And finally, (3.5) becomes

ˆ σ2(βi) = h_(1−R2 y) P i(Yi− ¯Y )2 n−k−1 i (1 − R_i2)P x2 i = h_(1−R2 i) P i(Yi− ¯Y )2 n−k−1 i P x2 i × 1 (1 − R2 i) = h_(1−R2 i) P i(Yi− ¯Y )2 n−k−1 i P x2 i × VIF

VIF (Variance Inflation Factor), quantifies the level of multicollinearity. In the litera-ture there is no consensus on how large the Variance Inflation Factor is allowed to be. Greene (1993) states that values exceeding ten indicate a problem of multicollinearity, while O’brien (2007) states the value of five. We test the most restrictive rule of thumb, a Variance Inflation Factor of five.

Distinguishability

After eliminating the variables that cause multicollinearity, we estimate the model for the whole time period 2015), before we estimate the model for the pre-crisis (2006-2007), crisis (2008-2009) and post-crisis (2010-2015) periods.

One could argue that the logit model that Olson and Zoubi (2008) and Khediri, Charfeddine, and Youssef (2015) propose is not suitable to distinguish between Islamic and conventional banks out-of-sample and that it can only be used within the sample. Whenever this is the case, the model can merely be used as a descriptive model and not as a model to distinguish between banks. To be able to tell whether the model can be used for distinguishing or not, we perform a within sample and an out-of-sample test considering other Islamic and conventional banks in the GCC region. Since the focus of this thesis lies on the post crisis period, we test the model for 2010 - 2015. We add com-parable (in terms of size and region) banks to the data set and we split the data set in two subsets. One data set is the training data set containing the original data and the other data set is the test set. Thereafter, we fit a logit model on the training set and we use that model to predict whether a bank is Islamic or conventional in the test set. These predicted probabilities are then compared to the actual banking types to conclude whether or not the logit model can be used out-of-sample to distinguish between both banking systems.

Chapter 3. Data and methodology 24

3.4.2 Financial ratio analysis

This analysis can be used to conclude whether there is evidence to believe that certain financial ratios are different between Islamic banks and conventional banks. It is used to answer the second subquestion: How did the conventional and Islamic banks in the sample recover from the 2008 financial crisis in terms of profitability, liquidity, risk and asset-quality? For this analysis, we conduct a univariate analysis following Olson and Zoubi (2008) and Khediri, Charfeddine, and Youssef (2015). The reason that we choose this technique is that it is a well established technique that is sufficient to answer our research question. We assume unequal variances since we observe differences in the variances of the finan-cial ratios (refer Appendix A). The test statistic is given by:

t = x1− x2 ps2

1/n1+ s22/n2

In this test statistic, x1, s21, n1and x2, s22, n2are the means, variances, and number of

obser-vations of the Islamic banks (group 1) and the conventional banks (group 2) respectively. The degrees of freedom are given by:

df = (s

2

1/n1+ s22/n2)2

(s2₁/n1)2/(n1− 1) + (s2₂/n2)2/(n2− 1)

The analysis is done for four different time periods. First of all, we perform an analysis for the overall period. Next, we divide the total period in three subsets, pre-crisis (2006-2007), crisis (2008-2009) and post-crisis (2010-2015). Since we observe the financial ratios of the same banks over time, the financial ratios of each bank are correlated with the financial ratios of the same bank in different time periods. This gives the problem of stationarity, which implies that we cannot fully rely on the t-statistics and the p-values of this test and we need to be careful with our conclusions. However, we still think that this method is useful for answering our research question, since it gives an idea whether the differences in means are increasing or decreasing.

3.4.3 Frontier analysis

As stated before, there are two main techniques to measure banking efficiency: data en-velopment analysis (DEA) and stochastic frontier analysis (SFA). These techniques are used to answer the third research question: Which banking system operates more efficient, the Islamic or the conventional banking system? Both techniques are used to assess the cost and profit efficiencies of both banking systems. Figure 3.5 provides an overview of the mod-els discussed in this section. The nodes in the first level describe the types of efficiency we incorporate in this thesis. The second level indicates the different methods and the third level shows the functional forms that are used for the parametric stochastic frontier analysis. The last level shows different methods to deal with negative profits.

Chapter 3. Data and methodology 25

FIGURE3.5: Overview of the models used for examining efficiency of both banking systems Efficiency Cost efficiency DEA SFA Cobb-Douglas Translog Profit efficiency DEA SFA Cobb-Douglas NPI Rescaled Translog NPI Rescaled

Data envelopment analysis

This nonparametric technique is used to differentiate between the efficiency of Islamic and conventional banks. Yudistira (2003) uses the same technique to measure the per-formance of eighteen Islamic banks over the period 1997 - 2000. In the literature, Farrell (1957) and Charnes, Cooper, and Rhodes (1978) are considered as the founders of DEA.

In the sample, every bank i is a decision making unit (DMU) in year t. Based on weighted input quantities, output quantities, input prices and output prices, every bank obtains a score. For cost efficiency, this score is between zero and one, since it is assumed that a bank always incurs positive cost and for profit efficiency, this score is between mi-nus infinity and one, since it is assumed that the profit of a bank can be negative. This score displays the efficiency of a bank, where one represents that a bank is fully efficient relative to the other banks in the sample and zero, or a negative score, implies that a bank is totally not efficient. These scores are obtained by solving a linear programming prob-lem. The outcomes display relative instead of absolute efficiency. In this context, relative efficiency means that the efficiency scores are relative to the other banks in the sample and the efficiency scores cannot be used to compare the efficiency of banks between dif-ferent researches with difdif-ferent samples. In Table 3.9, the variables for the DEA model are defined.

Chapter 3. Data and methodology 26

TABLE3.9: Introduction variables data envelopment analysis

Variable Meaning Conditions

CEit cost efficiency for bank i in year t ∈ (0, 1]

P Eit profit efficiency for bank i in year t ∈ (−∞, 1]

xjit input quantity variable j for bank i in year t ∈ R+

x∗_jit cost minimizing input quantity variable j for bank i in year t ∈ R+

yrit output quantity variable r for bank i in year t ∈ R+

y_rit∗ revenue maximizing output quantity variable r for bank i year t ∈ R+

pjit input price variable j for bank i in year t ∈ R+

qrit output price variable r for bank i in year t ∈ R+

λit weight for bank i in year t ∈ [0, 1]

Cost efficiency

We use Charnes, Cooper, and Rhodes (1978) and Coelli et al. (2005) as a basis to discuss the DEA model for cost and profit efficiency. The model for cost efficiency they use for DMU i starts with the following linear programming model:

min x∗ jit m X j=1 pjitx∗jit (3.6) s.t. n X b=1 xjbtλbt≤ x∗jit (3.7) n X b=1 yrbtλbt≥ yrit (3.8) n X b=1 λbt≤ 1 (3.9) with b = 1, ..., n i = 1, ..., n j = 1, ..., m r = 1, ..., s

This model is calculated for every year t ∈ (2006, ..., 2015). Furthermore we have n = 30 banks, m = 3 input variables and s = 3 output variables. The linear programming problem is minimized with respect to x∗_jit (Equation (3.6)) such that the weighted sum of the input quantities of the banks is lower than the optimal input quantity (Equation (3.7)). At the same time, the weighted sum of the output quantities of the other banks must exceed or be equal to the bank being evaluated (Equation (3.8)), given that the sum of the weights does not exceed one (Equation (3.9)). Then, the optimal cost scalar Pm

j=1pjitx ∗

jit is compared with the realized cost scalar

Pm

j=1pjitxjit for each DMU i in

each year t, to obtain the cost efficiency CEitfor bank i in year t.

CEit= Pm j=1pjitx∗jit Pm j=1pjitxjit (3.10)

Chapter 3. Data and methodology 27

Profit efficiency

For the profit efficiency model, we subtract the cost scalar from the revenue scalar for DMU i. The model is given by:

max x∗_jit,y∗_jit s X r=1 qrityrit∗ − m X j=1 pjitx∗jit (3.11) s.t. n X b=1 yrbtλbt ≥ yrit∗ (3.12) n X b=1 xjbtλbt≤ x∗jit n X b=1 λbt≤ 1 with b = 1, ..., n i = 1, ..., n j = 1, ..., m r = 1, ..., s

The idea behind this linear programming problem is the same as the linear programming problem of the cost efficiency. In Equation (3.11), the revenue minus cost is maximized with respect to the input and output quantities. The conditions are the same as in the cost efficiency case, with the difference that the weighted sum of the output quantities of the other banks must exceed or be equal to the optimal output quantity (Equation (3.12)). In the cost efficiency case, the output quantity is not optimized. Then, we com-pare the optimal profit vectorPs

r=1qrity ∗ rit− Pm j=1pjitx ∗

jitwith the realized profit vector

Ps

r=1qrityrit−Pmj=1pjitxjitfor DMU i to obtain the profit efficiency P Eit for bank i in

year t. P Eit= Ps r=1qrityrit− Pm j=1pjitxjit Ps r=1qrityrit∗ − Pm j=1pjitx∗jit (3.13)