Impact of Islamic Finance on the Great Moderation- How it affects volatility of the business cycle

55
Student no: 620029286 1 Impact of Islamic Finance on the Great Moderation: how it affects volatility of the business cycle. Submitted by Prashina Gobind Samtani to the University of Exeter as a dissertation for the Master of Science (MSc) in Economics September 2013 I certify that all material in this dissertation, which is not my own work has been identified and that no material is included for which, a degree has previously been conferred upon me. Prashina Gobind Samtani

Transcript of Impact of Islamic Finance on the Great Moderation- How it affects volatility of the business cycle

Student no: 620029286 1

Impact of Islamic Finance

on the Great Moderation:

how it affects volatility of

the business cycle.

Submitted by Prashina Gobind Samtani to the University of Exeter as a dissertation

for the Master of Science (MSc) in Economics September 2013

I certify that all material in this dissertation, which is not my own work has been

identified and that no material is included for which, a degree has previously been

conferred upon me.

Prashina Gobind Samtani

Student no: 620029286 2

Acknowledgements:

This dissertation “Impact of Islamic Finance on the Great Moderation: how it affects

volatility of the business cycle” is intended to serve the purpose of the authors’ final

term research paper in her postgraduate degree in the field of Economics.

Firstly, I would like to express my heartfelt gratitude to my supervisor Dr. Dudley

Cooke for his support and guidance throughout the journey of my research. His belief

has not only enabled me to discover this challenging area of economics but also

developed a better understanding of the econometric methods required to tackle this

challenge.

I am also thankful to my economics classmates, who not only have been able to boost

me whenever I was striving towards my goal but they also ease me with some

suggestions to move forward.

Last, but not least, to my parents, my sister and my nephew, who have been my utmost

inspiration and the pillars to keep me going till completion. I am most grateful to the

Lord Almighty for making all these possible.

Student no: 620029286 3

Abstract:

The Niche Islamic Finance has been growing considerably in the recent years. It is

useful to study the impact of this growth on the business cycle. GDP volatility is the

best proxy for the fluctuations that alter the business cycle. The Great Moderation

period coincides with the boost on the growth of Islamic Finance. The paper will

therefore model the relation between the two events. The emphasis would be on the

different banking systems affecting the low volatility of the business cycle that marks

the concept of Great Moderation. Panel regressions across 116 sample countries are

then run to experiment the hypothesis proposed. The panel was sub-divided to test

whether the percentage of Islamic Banking system in the country has an impact on the

volatility tested.

The results indicated that the presence of Islamic Banking systems contribute

significantly in explaining the low volatility during the period of interest. However, the

percentage of concentration of Islamic Banking system in each country does not

significantly explain this phenomenon.

Keywords: Islamic Finance, Business Cycle, Great Moderation, GDP volatility and

Panel regressions.

Student no: 620029286 4

List of Contents:

1. Introduction Page 6

1.1 Concept Page 6

1.2 Background History Page 6

1.3 Significance Page 9

2. Literature Review Page 9

3. Research Design Page 15

3.1 Sample Data Description Page 15

3.2 Dependent Variable Page 15

3.3 Independent Variables Page 15

3.3.1 Oil Volatility Page 15

3.3.2 Islamic Finance Dummy Page 16

3.3.3 Developing country Dummy Page 16

3.3.4 Investment Page 19

3.3.5 Inflation Volatility Page 19

3.3.6 Real Interest Rate Page 19

3.3.7 Credit Page 19

3.3.8 Trade Openness Page 19

3.3.9 Islamic banking concentration Page 20

4. Statistical Methodology Page 20

4.1 Panel Regression Page 20

4.1.1 Cross correlation between the variables Page 21

4.1.2 Descriptive Statistics Page 22

4.1.3 Variance Inflation Factor test Page 23

4.2 Sub-panel regression Page 23

4.3 Method of Estimation Page 27

5. Results Page 31

5.1 Panel Regression Page 31

5.2 Sub-panel Regression Page 33

5.3 Panel Validity Tests Page 34

6. Discussion Page 41

7. Limitations and Further Research Page 43

8. Conclusion Page 45

9. Appendices Page 46

10. Glossary Page 50

11. References Page 51

Student no: 620029286 5

List of Figures and Tables:

Table 1- Major Differences between Islamic & Conventional Banking System. ............... 8

Table 2 - Summary of Literature Review and Usefulness to the Research ..................... 14

Table 3 - Relation between Developing Dummy and Islamic Finance Dummy ............... 18

Table 4- Sub-division of main regression panel ................................................................ 21

Table 5- Correlation matrix of Regression variables ........................................................ 22

Table 6- Description statistics of Regression Variables .................................................... 23

Table 7- Variance Inflation Factor test ............................................................................. 23

Table 8- Sub-division of sub-panel regression ................................................................... 26

Table 9- Computation of transformation parameter in Random Effects Model .............. 29

Table 10- Hausman test ..................................................................................................... 30

Table 11- Panel Regression results .................................................................................... 31

Table 12- Sub-panel Regression results ............................................................................ 33

Table 13- Model Selection test A ........................................................................................ 36

Table 14- Model Selection test B ........................................................................................ 37

Table 15- Serial Correlation validity tests ........................................................................ 38

Table 16- Heteroskedasticity validity test ......................................................................... 39

Table 17- Transformed final model robust to heteroskedasticity and correlation ........... 40

Figure 1-Period of Great Moderation presented by Annual GDP growth over time ......... 6

Figure 2-Diffusion timeline of Islamic Banking System ..................................................... 7

Figure 3- Phase Diagram of Investment............................................................................ 12

Figure 4-Changes of Oil Volatility over time ..................................................................... 16

Figure 5- Number of Islamic Banks across the Globe ....................................................... 25

Figure 6- Volatility of 89 countries over time………………………………………………….26

Figure 7- Volatility of 23 countries over time………………………………………………….26

Figure 8- Flowchart of tests to select appropriate method of estimation………………….35

Student no: 620029286 6

1. Introduction

1.1 Concept

Great Moderation is a concept that arises around 1985 when the periods of low

volatility in the business cycle begun. Many empirical literatures have investigated the

occurrence of this pattern. Gradually, the Great Moderation lost its publicity due to the

occurrence of financial bubble crisis and most of the research shifted towards this area.

It is of no surprise that Financial Intermediaries play a key role in economic growth as

stated by the report in IMF (2000, p.164). Coincidently, a niche finance market known

as Islamic finance started growing in the US and European region around the same

period. The banking systems under this market must operate within the framework of

the religion, based on Quran and Sunnah. The law that binds this finance market is

popularly known as the Sharia Law. The main difference between the banking system

in the Islamic manner and the conventional is the prohibition of Riba, which is the

interest imposed on money itself. The banking framework used under Islamic manner

is known as Profit-Loss Sharing (PLS) mechanism.

1.2 Background History

The period of low volatility of business cycle has been around from 1985-2007, where

the crisis put an end to it.

Source: National Income and Product Accounts

Figure 1-Period of Great Moderation presented by Annual GDP growth over

time

Student no: 620029286 7

On the other hand, the concepts adopted by the Islamic finance system are not new

since it rooted from the barter system. But, the modern Islamic Banking system is

considered a recent phenomenon. In Imam and Kpodar (2010) report, they tracked the

recent establishment of this banking system from the 1960s appearance. Murdoch

further discusses the timeline in a Reuters article in 2009, which is summarized in the

flowchart below.

Source: Reuters, Sole (2007) and Author’s compilations

Figure 2-Diffusion timeline of Islamic Banking System

Updating the timeline of the Reuters article to a more current research, Sole (2007) has

stated that 300 Islamic Financial institutions and over 250 mutual funds have

complied with the Islamic principles. It was also found that the growth rates of the

industry have been impressive and about 10-15 percent per year. In the future, this

trend is likely to linger.

1950s-1960s

•First experiment done in Pakistan and Indian Sub-continent.

•Malaysia and Egypt pioneer ventures using the concept in 1960s.

1975

•First commercial bank in Dubai was established.

•The Islamic Development Bank opens in Jeddah, Saudi Arabia.

1979 •Pakistan pioneered to Islamize the banking system.

1983

•Malaysia established first official sharia-compliant Bank.

•Sudan develops the dual banking system (Conventional along with Islamic).

1984 •Iran begin to Islamize the banking system.

1985 •50 Financial Institutions offered Islamic Financial products around the Globe.

1990s •Other major banks followed offering the Islamic financial products

by 2000

•About 200 Islamic Financial institutions developed over $8 billion in capital. The countries range over Middle East, Southeast Asia, Africa, Europe and America

by 2007

•300 Islamic Financial Institutions and 250 mutual funds complied with the Islamic Principles around the Globe

Student no: 620029286 8

Therefore, the substantial diffusion of Islamic Banking throughout the past 40 years

cannot be overlooked. As Bashir (2007) has indicated in his paper, Islamic Banks have

successfully entered international markets. He stated ‘Islamic Banking is one of the

fastest growing banking industry segments worldwide, with an annual growth rate of

17 percent and asset base of US$ 23 billion for the past several years’ (p. 3). At first the

entry of the Islamic Banks were motivated by the need of the Muslims who refuse to

put their money in the Conventional Banks. Following on, the spread became more

prominent after the success of some Islamic Banks drew the attention of the Asian and

European countries. Some of the banks were then set up in countries that have

significant Muslim minorities in order to seize profit-making opportunities. It is

therefore important to acknowledge its impact to those countries.

Before we assess the effect of Islamic Banking system on the business cycle, it is

important to familiarize the reader with the operation of the system. They operate in a

manner different from the conventional banking system in various characteristics,

which are shown in Table 1.

Source: Adib1 (n.d.), Kayed and Hassan (2011) and Author’s compilation.

Table 1- Major Differences between Islamic & Conventional Banking System.

1 The complete URL of the webpage where the information is compiled in the table is per:

http://www.adib.ae/understanding-islamic-banking-0.

Characteristics Islamic Banking System Conventional Banking System

Framework Follows Sharia laws Follows secular banking laws

Prohibition of Gharar

Transactions that involve a lot uncertainty are not allowed. Example would be

options trading.

There is no such prohibition in derivatives trading of any forms.

Profit and Loss Sharing

Main paradigm used in the system. Banks also loses money if the investment fails.

One of the symmetric profit sharing is the Musharakah contract.

No such principle, the investor bears the losses and enjoys the gains as a

whole. The Banks are entitled to fixed fees stipulated in the contract.

Prohibition of Riba

Interest on money itself is prohibited. The banks gains by applying a certain mark up or fees on the financing resource provided

to customers.

There is no such prohibition, as long as the customers meet the

application requirement, interest will be charged according to the credit

standing.

Industry

No trading activities on certain industries that deals with alcohol, drugs, gambling and others that are considered unlawful

under Sharia law.

There is no such prohibition on any industry trading activities.

Prohibition of Maysir

There is a prohibition on transactions that involve speculation.

No such prohibition on the assets traded in the conventional system.

Student no: 620029286 9

1.3 Significance

The question is whether the existence of Islamic Banking system could help explain the

research debate about the sources of Great Moderation.

The main idea of this research is to test the significance of the relationship between the

two events. The implication of testing the relation is to take into account the effect of

Islamic finance on explaining the volatility of the business cycle for future policy

making activity. The volatility is important in policy making because the greater the

volatility, the greater the uncertainty segment embedded on it. If the sources of

stabilizing volatility are found, more robust policies are implemented and fewer shocks

will affect the system.

This research paper will be divided into several sections. Section 2 will enlighten the

literature regarding the topic of interest. I will design the research in Section 3,

describe the method in Section 4 and present the results in 5. The results will be

discussed in Section 6. Suggestions for further research will be illustrated in Section 7.

Lastly, the conclusion of the paper will be deduced in Section 8.

2. Literature Review

Various studies have been conducted in regard to the sources of declining output

volatility and existence of Great Moderation. Although, Gali and Gambetti (2009)

agreed that there is some consensus among the studies on the existence and the period

of the Great Moderation, they also consented that there are controversies on its

interpretation. Creal et al. (2010) also affirmed that the causes of Great Moderation are

still much open to research. Creal et al. (2010) stated that Burns and Mitchell (1946) is

the key inspiration of other research in business cycle due to its immense contribution

in policy and decision-making.

Firstly, Kent et al. (2005) did a panel data study on 20 Organization for Economic Co-

operation and Development (OECD) countries and investigated the sources of Great

Moderation. There are four factors that have been proposed in the paper. Behavior of

the different components of GDP, policy efficiencies, structural changes in the industry

or just plain good luck. The good luck here is just an absence of bad luck for example oil

price shocks, which was absent during the Great Moderation period. They ran panel

regression from the period of 1973 until 2003, computed the GDP volatility on a rolling

window of 5-years and regress that as the dependent variable to the structural

indicators.

Student no: 620029286 10

Their main regression is as such:

Where is the standard deviation of annual growth of real GDP for country

is a vector of direct structural indicators at previous period such as product

market regulations and monetary policy regime.

is a vector of indirect structural indicators at the same period such as days

lost to labour disputes, trade openness, financial liberalization, inflation volatility

and fiscal policy volatility.

is a vector of other possible explanatory variable, in this case they have

included oil volatility since it is constant at a period of time for all the countries

regressed.

After running the regression, the results did not coincide with the literature and most

of the structural indicators appear to be insignificant in their results. They stipulated

that one reason behind this could be that some trend data may not have been taken

into account during 1973-1985, which might differs much with the trend of the rest of

the dataset. Using a single country was suggested as a control. Therefore, United

States dataset was proposed as a ‘locomotive’ in their paper.

Additionally, Gali and Gambetti (2009) believed that their paper has been able to shed

some light on the advantages of introducing the reasons of the existence of the Great

Moderation. They used volatility changes by computing standard deviations of output

for the pre and post-1984 data and compared the difference in volatilities using US

quarterly data from 1948-Q1 up to 2005-Q4. They utilized the Real Business Cycle

(RBC) Theory and applied Vector-Auto-regression (VAR) to examine the period. Their

paper focused on identifying the role played by different type of shocks that affects the

output volatility in the period of interest. In their research, they found that

Investment-type technology shocks and Non-technology shocks are the main rationale

behind the occurrence of the Great Moderation. This finding confirms the results of

Justiniano and Primiceri (2008) and Fisher (2006) who identified that Investment-

related shocks dominate the reasoning behind the decline in volatility of output growth.

In addition, Justiniano and Primiceri (2008) findings also suggest that investment

shocks are the triggering factors for other key macroeconomic variables changes in the

Student no: 620029286 11

cycle. The only thing that differs in their research was a larger dataset and the use of

Dynamic Stochastic General Equilibrium (DSGE) model as a method of estimation.

Following on, Neumeyer and Perri (2005) test some variables that move the business

cycle and found that real interest rates are countercyclical for emerging economies and

acyclical for developed economies. They performed their experiments in a

computational manner and focused on data from Argentina for the period 1983-2001.

Their conclusion was based on 5 developing countries and 5 other developed economies

under study. Moreover, they found that the extent of country risk in developing

countries affects the degree of countercyclical movements of real interest rates to the

cycle.

Then, Benati and Surico (2009) did a similar study focusing on United States, but used

different methods to investigate which of the sources in the list contributed to the

existence of Great Moderation. They used Structural VAR method to test the validity of

good luck reasoning. They used data of pre- and post-Great Moderation and found that

the period of interest is more volatile than the other two segments. Subsequently, they

used DSGE model that suggests that systematic component changes in GDP and the

low variance of investment are the main justification for Great Moderation to reign.

The literature focus then shifted to the investment reasoning. Salahuddin and Islam

(2008, p.21) have stated ‘Investment is the nucleus of an economy’. According to them,

investment plays a vital role in economic growth. So, in order to prove their theory of

accelerated growth, which can be caused by investment, they did a panel data study on

97 randomly selected developing countries on investment that includes both public and

private as the dependent variable from 1973 to 2002. The study done in that paper

confirms the significance of investment to empirics of growth and therefore influences

the economy to a great extent.

Building on the importance of investment, Chishti (1985) concluded that the amplitude

of all phases in business cycle under an Islamic economy would be smaller than in a

conventional economy. The main rationale that follows the argument is the instability

that is transmitted through the investment process that differs under the two systems.

He used Minsky Investment theory to reach the conclusion stipulated before. According

to the latter theory, the fragility of the financial system depends on the relation

between its financing decision (cash commitments) and their profits (cash flow). The

Student no: 620029286 12

extent of investment therefore depends on the two variables. He used a phase diagram

to explain the relation.

Source: Chisthi (1985)

Figure 3- Phase Diagram of Investment

The figure shows that investment grows as external financing needs increase and

becomes more stringent as represented by the f=0 line. On the other hand, when the

cash flows increases there is more scope for investment but at the same time limited to

its savings. Therefore, the need to rely on external financing surfaces with time, as

represented by the g=0 line. As a result, a balance around the circle represented in the

central segment of the graph has to be maintained.

Moreover, he postulated in his paper that under Islamic Financing scheme, investment

focused only on productive purpose therefore creating a link between the two variables.

The bond between payment commitments and cash flows helps to move both variables

together, hence forming stability in the system. The gap is thus minimized greatly

under the Islamic system at micro-investment level.

In this paper, we will examine the effect at macro level and the relation of the two

events from an elevated view. Imam and Kpodar (2010) investigated the diffusion of

Islamic Banking at the macro level using cross-country Poisson distribution regression

during the period 1992-2006. The sample consists of 117 countries and their hypothesis

was to find out the factors that help the spread of Islamic Banking during the years.

Student no: 620029286 13

Some of the factors that found to be explanatory of the hypothesis are the share of

Muslims in the country, income per capita, and the respective country’s distance to

Major Islamic Centres. Moreover, higher interest rates impact negatively on the

diffusion of Islamic Banking. Evolution of oil prices over the years also serves as a

great explanation for the spread of Islamic Banking. The stability of the economy and

its trade with the Middle East are also conducive in its contribution to the variable

tested. The highlight of the findings in this paper suggests that the September 2011

attacks in United States has nothing to do with the growth of Islamic Banking,

escalating oil prices are indeed the real driver.

Beck et al. (2013) compared the Conventional Banking system with the Islamic

Banking system in terms of cost efficiency, asset quality, stability and business

orientation. Using Bankscope’s database, they sample the largest 100 banks in terms of

assets within a country for the period of 1995-2007, pre financial crisis. After

elimination of outliers from the sample, they tested two different sample datasets, one

with all the countries and another with only countries that practice both Islamic and

Conventional banking Systems. Then, they ran two different panel regressions and the

empirical findings suggest very little or insignificant differences between the two

systems in terms of the areas discussed. The main difference between the two sample

data is the second smaller sample includes a variable of percentage of Islamic Banks as

compared to Conventional Banks in the countries that practice Islamic Banking. The

second sample is referred to Islamic countries. However, with the second panel dataset,

the estimates indicate that those countries with a larger concentration of Islamic

Banks are more cost effective but are less stable than those of lower concentration.

Lastly, I have summarized the literature above in a tabulated format and added the

usefulness for this research.

Student no: 620029286 14

Source: Literature Review section and Author’s summary

Study object

Authors Published

year Period of study Method Findings

Useful for Research

United States

Gali and Gambetti

2009 1948-2005

Compared volatility changes between pre and post Great Moderation period using VAR method with the application of RBC theory.

Investment-type technology shocks are the main rationale under Great Moderation.

Focus on Investment

United States

Creal et al. 2010 1953-2007 Stochastic volatility is assumed in the model to predict business cycle indicator that vary over time.

Structural breaks occur pre and post 1984, concludes Great Moderation is open to research.

Research motivation

20 OECD countries

Kent et al. 2005 1973-2003

Computed GDP volatility by rolling window of 5 years and regressed as dependent variable. Panel regression for the period.

Trend in output volatility is the main reason for decline in magnitude of global shocks.

Computation of GDP volatility and Inflation volatility is an important variable.

United States

Justiniano and

Primiceri 2008 1954-2004

Using DSGE model to investigate the sources of shifts of volatility in the US during the postwar period.

Investment shocks is key factor in business cycle changes

Focus on Investment

10 Countries

Neumeyer and Perri

2005 1983-2001 Computational experiments to evaluate the role played by interest rate shocks to the economy.

Real interest rates are countercyclical for developing and acyclical to developed economies.

Real interest rate as a variable and Developing country dummy.

United States

Benati and Surico

2009

2000 simulations pre Oct 1979 and post Volcker.

Structural VAR and DSGE model to investigates the sources of Great Moderation existence.

Systematic component changes in GDP and low variance of investment are the main reasons.

Focus on Investment

97 Developing countries

Salahuddin and Islam

2008 1973-2002 Investment as the dependent variable Confirms that Investment is the nucleus of an economy

Focus on Investment

Qualitative Model for

firms Chishti 1985 Not applicable

Using Investment theories to develop two differential equations. Formulated relation between financing and investment decisions at a micro-level.

Amplitude of all phases in business cycle under Islamic economy is smaller as compared to the conventional system

Country level of investment is an aggregate of micro-firm behaviors.

117 countries

Imam and Kpodar

2010 1992-2006 Using Poisson distribution method to regress factors that might affect the growth of Islamic Banking.

Oil prices are real driver of Islamic Banking growth

Oil Volatility is important to be taken into account.

100 largest banks

Beck et al. 2013 1995-2007 Panel regression of cross-country regressions of two different samples.

Islamic Banks are more cost effective but less stable.

Method of sub-panels are adapted

Table 2 - Summary of Literature Review and Usefulness to the Research

Student no: 620029286 15

3. Research Design

3.1 Sample Data Description

The sample consists of 116 countries randomly selected from the pool of 214 countries

available during the period of 1985-2007. The countries selected in this sample are

presented in Appendix 9.1. The period selected corresponds to the Great Moderation

concept introduced to explained the period of low volatility in the business cycle before

the financial crisis breakout.

3.2 Dependent Variable

The Dependent variable to explain the volatility in the business cycle is closely proxied

by Gross Domestic Product (GDP) data of the countries. Therefore, I have extracted the

annual growth of GDP per capita data from those 116 countries. Further, the standard

deviation is computed over five-year period in a similar fashion as in Kent et al. (2005),

which result in the truncation of the data starting now from 1989 onwards. This

method is also justified in IMF report (2012, p. 167) where the standard deviation is

found by looking at a backward rolling 5-year window. The resultant variable is then

termed as the GDP volatility. The data is taken from the World Bank, World

Development Indicators (WDI) for this variable.

3.3 Independent Variables

The independent variables have been extracted from various sources due to the vast

nature of the variables to be included in the model. The next few subsections will

further describe the various sources.

3.3.1 Oil Volatility

The variable here is proxied by the price of crude oil. World crude oil prices are

established in relation to three market-traded benchmarks (WTI, Brent and Dubai) as

indicated by the International Energy Agency. The nominal spot prices here is

benchmarked by the Dubai market and expressed in US dollars per barrel of oil making

all countries to state the same price at each period of time. This data is extracted from

OECD statistics. The variable evolves over time as graphed below:

Student no: 620029286 16

Figure 4-Changes of Oil Volatility over time

Here, the volatility is computed in a similar manner by taking standard deviation over

five-year period. Many countries used the oil prices as an opportunity to introduce

Islamic Banking system as indicated in Bashir (2007) and Imam and Kpodar (2010),

which explains that oil price fluctuations will be an important variable in research. The

variable is useful to observe the effect of productivity shocks to the country’s GDP in

terms of its fluctuations.

3.3.2 Islamic Finance Dummy

The dummy is introduced to distinguish those countries that practice Islamic finance

regardless of the percentage of its participation in the area. This is done by taking into

account the existence of at least one Islamic bank in the country examined. The

rationale behind the inclusion of this variable is to test the theory proposed in Chisthi

(1985), which states that the amplitude of the business cycle under an Islamic Economy

is lower than in a Conventional system. Those countries that practice Islamic Banking

therefore take the value 1 and 0 otherwise. The list of the countries that practice

Islamic banking currently, have been extracted from the Association of Islamic

Banking Institutions Malaysia website.

3.3.3 Developing country Dummy

The dummy here introduced is to substantiate our literature that summarizes that

developed countries behave in a different manner towards the business cycle variation

Student no: 620029286 17

as compared to those that is still emerging. This has been backed up by Neumeyer and

Perri (2005) research findings.

The list of developing countries is extracted from International Statistical Institute

website, which is effective for the whole year 2013. Those that are developing therefore

take the value 1 and 0 for the developed ones. The site defined that the Countries with

Gross National Income (GNI) per capita less than US $ 11,905 are considered as

developing countries in the list.

We can represent the two dummy variables in a graphical view in order to have an

indication of the sampled countries characteristic2 as per below.

We can see from the table of 116 countries below, it is quite rare to find developed

countries that practice Islamic Banking.

In the table on the next page, we can see that the majority of the countries lie in the

bottom-right box of the matrix. This means 51 countries out of the list of developing

countries do not practice Islamic Finance. The list of the countries in the other

categories is represented in the list in full on the following page.

2 The characteristic here represents inherent feature of the country, that is, the developing characteristic

and an existence of the Islamic Banking system.

Student no: 620029286 18

Table 3 - Relation between Developing Dummy and Islamic Finance Dummy

3 Countries that practice Islamic Banking take the value 1 and 0 otherwise. 4 Developing countries takes the value 1 and 0 for developed ones.

3

1

Brunei Darussalam

Cyprus

Israel

Singapore

Switzerland

United Kingdom

United States

Algeria

Azerbaijan

Bangladesh

Egypt, Arab Rep.

Ethiopia

Gambia, The

Indonesia

Iran, Islamic Rep.

Jordan

Kazakhstan

Kenya

Malaysia

Mauritania

Mauritius

Pakistan

Philippines

Russian Fed

Senegal

South Africa

Sri Lanka

Sudan

Syrian, Arab Rep.

Tanzania

Thailand

Tunisia

Turkey

Uganda

Yemen, Rep.

0

Australia

Austria

Bahamas

Barbados

Belgium

Canada

Czech Republic

Denmark

Estonia

Finland

France

Germany

Greece

Hong Kong

Hungary

Iceland

Italy

Japan

Luxembourg

Macao

Malta

Netherlands

New Zealand

Norway

Poland

Portugal

Slovak Republic

Slovenia

Spain

Sweden

Trinidad &Tobago

Albania

Argentina

Armenia

Belarus

Belize

Benin

Bolivia

Botswana

Brazil

Bulgaria

Burkina Faso

Cambodia

Cameroon

Cape Verde

Chad

China

Congo Dem.Rep.

Congo, Rep.

Costa Rica

Cote d’Ivoire

Dominican Rep

Ecuador

El Salvador

Gabon

Guatemala

India

Korea, Rep

Kyrgyz Republic

Latvia

Lesotho

Lithuania

Macedonia, FYR

Madagascar

Mali

Mexico

Moldova

Morocco

Mozambique

Myanmar

Panama

Papua New Guinea

Paraguay

Peru

Romania

Rwanda

Swaziland

Togo

Ukraine

Uruguay

Zambia

0 1

4 Developing Dummy

Student no: 620029286 19

3.3.4 Investment

Gali and Gambetti (2009), Justiniano and Primiceri (2008), Benati and Surico (2009)

and Salahuddin and Islam (2008) would agree that Investment is one of the chief

explanatory variables to explain the fluctuations of the business cycle. The variable

taken to represent Investment is known as Gross Capital Formation in the database

World Bank (WDI). It represents the outlays on any additions to the fixed assets of the

economy and changes in the level of inventories as expounded in the variable catalog.

The variable is represented as percentage of Gross Domestic Product (GDP). The

inclusion of this variable is necessary to measure the country’s GDP growth in a more

comprehensive manner.

3.3.5 Inflation Volatility

The variable for inflation is extracted from World Bank (WDI). The variable involved

was measured by the Consumer Price Index (CPI) since it reflects the direct impact of

acquisition of basket of the goods and services that will cost the consumer (IMF, 2012),

unlike the variable measured by GDP deflator affecting the economy as a whole. This

variable is an indirect indicator to measure the effectiveness of monetary policies

implemented in a particular country and its stability. Kent et al. (2005) agreed that

this variable is indeed one of the main structural change indicators of the business

cycle volatility.

3.3.6 Real Interest Rate

As suggested by Neumeyer and Perri (2005), real interest rates’ behavior is important

to the business cycle movements. This variable extracted from the World Bank (WDI) is

the lending interest rate adjusted for inflation.

3.3.7 Credit

This variable plays a vital role to explain how financial institutions affect growth

volatility, which elucidates the financial resources provided by the banks to the private

sector as a percentage of GDP. Easterly et al. (2000) confirmed the role stated, taken

the variable to measure the financial depth in their research. This variable that was

extracted from World Bank, the Global Financial Development Database as Bank’s

private credit to GDP, was also used to measure the extent of financial system

development in the country and their contributions towards the changes in GDP.

3.3.8 Trade Openness

Student no: 620029286 20

This variable is closely proxied by the sum of exports and imports across the countries

as a percentage of their GDP also supported in IMF (2012, p.167). The data is thus

obtained from the World Bank (WDI) database as well. The usefulness of this variable

is undoubtedly apparent due to its direct relation to the computation of any country’s

GDP.

3.3.9 Islamic banking concentration

This paper has adapted the Beck et al. (2013) method of panel data estimations, where

the percentage of Islamic Banking in Islamic countries is taken as a variable and

applied to a different dataset. The list of banks has been extracted from the Association

of Islamic Banking Institutions Malaysia site. Then, acknowledging the concentration

in each country, a percentage of total Islamic banks existing up to date as compared to

the whole pool is then computed.

4. Statistical Methodology

The method in my statistical research will be mainly focused on running a panel data

analysis on the variables explained across the countries in the earlier section. Stata5 is

the main software used to run the commands for this research. The method is largely

motivated by Kent et al. (2005) regression but adapted to suit the research focus of this

paper. The adaptations will be explained in this section in detail. All the variables are

extracted based on annual data during the period of the Great Moderation that is of

great interest in this paper. The main advantage in using panel data research is its

ability to embed variations both across countries and over the time period under study.

This is because engaging plain time-series method will neglect the possible correlation

across countries and cross-section methods will fail to take into account the

heterogeneity effect. As a result, the country-specific effects like the Islamic Finance

and developing country characteristic can be captured by using panel data since it

addresses the problem of biasness and inconsistencies in the coefficient estimates that

might occur under Ordinary Least Squares method.

4.1 Panel Regression

After refining the dataset due to missing values, the regression is finally run over the

period that ranges from 1991-2007. The regression equation is as follows:

5 The Stata log files have been compiled and are available upon request.

Student no: 620029286 21

Where:

= Standard deviation of annual growth of GDP per capita for country i

= Constant term

= Standard deviation of oil price per barrel in US dollars for period t.

= Dummy variable for existence of Islamic finance banking system for country i.

= Dummy variable for Developing country characteristic for country i.

= Vector of independent variables that affect GDP volatilities such as

Investment, Trade Openness, Credit, Real interest rate and Inflation volatility.

First, this panel will use a sample of 116 countries. Then, various regressions are going

to be run from which certain countries will be dropped due to missing values in the

data. Therefore, the sampled countries are adjusted to a set of 102 and 89 countries as

listed in Appendix 9.2 and 9.3 respectively. The variables regressed for the different

sample sizes are summarized by the � symbol in table 4.

Variables Panel-116 Panel-102 Panel-89

Developing � � �

Islamic Finance � � �

Investment � � �

Oil Volatility � � �

Inflation Volatility � � �

Trade � � �

Interest � � �

Private credit � � �

Table 4- Sub-division of main regression panel

There are a few statistical tests that are important prior to the usage of the data in the

panel regression. This is necessary in order to equip the reader with a rough indication

on the relationship between the variables. These few tests are performed based on 89

countries since some country data are lacking if the variables are to run together.

4.1.1 Cross correlation between the variables

This correlation matrix is computed in order to test whether the correlation amongst

the variables is significant at 5% level.

Student no: 620029286 22

GDP volatility Developing

Islamic Finance

Investment Oil Volatility Inflation Volatility

Trade Interest Private credit

GDP volatility 1.0000

Developing 0.2166*

(0.000) 1.0000

Islamic Finance

-0.1278* (0.0000)

0.1557* (0.0000)

1.0000

Investment 0.0759*

(0.0036) 0.0629*

(0.0000) -0.0330 (0.2047)

1.0000

Oil Volatility -0.1694* (0.0000)

0.0000 (1.0000)

0.0000 (1.0000)

0.1006* (0.0001)

1.0000

Inflation Volatility

0.2455* (0.0000)

0.1034* (0.0001)

-0.0751* (0.0039)

-0.0306 (0.2453)

-0.0852* (0.0011)

1.0000

Trade 0.0140

(0.5887) -0.1943* (0.0000)

-0.0193 (0.4532)

-0.0022 (0.9334)

0.1154* (0.0000)

-0.0522* (0.0455)

1.0000

Interest -0.0561* (0.0345)

0.0742* (0.0051)

-0.0423 (0.1104)

-0.0501 (0.0613)

-0.0709* (0.0074)

0.0208 (0.4341)

-0.0817* (0.0020)

1.0000

Private credit -0.2362* (0.0000)

-0.5992* (0.0000)

0.0899* (0.0006)

-0.1058* (0.0001)

0.1116* (0.0000)

-0.1318* (0.0000)

0.2671* (0.0000

-0.1172* (0.0000)

1.0000

Table 5- Correlation matrix of Regression variables

* Significant at 5% level

There is a linkage between correlation and multi-collinearity. High correlation may

increase the chances of multi-collinearity to arise. Although many of the correlations

are significant in the table above, the highest cross correlation shows between the

variable private credit and developing dummy (-0.5992). This indicates there is a strong

negative relationship between these two variables. The next highest significant value

in line is the relation between private credit and trade openness that is positively

related (0.2671). Although most of them are significantly correlated, the correlations

between the variables are not vey high at a glance. In section 4.3, some specification

tests will be further performed taken into account this indication of correlation.

However, it is advisable to run through the multi-collinearity test to ensure the

regression does not suffer from it.

4.1.2 Descriptive Statistics and Pattern

The table below is to summarize the statistics of the variables to be used for the

regression and to have a light indication on the variance of the factors being

considered.

Student no: 620029286 23

GDP

volatility Developing

Islamic Finance

Investment Oil

Volatility Inflation Volatility

Trade Interest Private credit

Mean 2.9709 1.6404 1.2809 5.7930 1.2975 38.7704 85.4847 7.7064 47.1607

Maximum 30.8131 2.0000 2.0000 223.0843 3.7817 3287.5730 436.9581 374.309 272.8089

Minimum 0.0520 1.0000 1.0000 -87.2000 0.3985 0.0421 0.3088 -91.7244 0.1152

Standard deviation

3.0064 0.4800 0.4496 15.9712 0.9824 257.6522 54.8147 15.9256 42.3844

Observation 1505 1513 1513 1474 1513 1474 1510 1424 1441

Table 6- Description statistics of Regression Variables

4.1.3 Variance Inflation Factor test

This test helps to identify any multi-collinearity that might occur between the

variables. This problem as explained by Verbeek (2004) occurs when a linear

relationship among the independent variables leads to unreliable regression estimates.

Variable VIF

Developing 1.69

Islamic Finance 1.11

Investment 1.04

Oil Volatility 1.06

Inflation Volatility 1.03

Trade 1.11

Interest 1.06

Private credit 1.73

Table 7- Variance Inflation Factor test

The test is computed by using the individual R2. The formula is as follows:

The threshold is either 5 or 10 depending on the sample size as indicated in Salahuddin

and Islam (2008). All the above results are below the level; therefore we can deduce

that the variables in the regression are free from multi-collinearity problem.

4.2 Sub-panel regression

This regression only includes those countries deduced from 4.1 regressions that

practice the Islamic banking system and refining the sample into those countries that

have complete dataset. This panel has been adapted from the paper done by Beck et al.

(2013) by concentrating the Islamic countries data only. The equation has therefore

been reformed to:

Student no: 620029286 24

Where:

= Percentage of concentration of Islamic banking in a country i as compared to

the total Islamic banks.

The figure presented below aims at depicting the number of Islamic banks in each

country across the globe. Only 49 countries display the presence of Islamic Banks. The

concentration in a particular country is then computed by calculating the proportion

between the number of Islamic Banks present in that country with respect to the total

number of Islamic Banks.

Student no: 620029286 25

Source: Association of Islamic Banking Institutions Malaysia, Author’s compilations

Figure 5- Number of Islamic Banks across the Globe

0

5

10

15

20

25

30

Number of Islamic Banks across countries

Student no: 620029286 26

The 49 countries shown in the figure above are not fully represented in the original 116

sampled countries. From Table 3, we can observe that the first row of the matrix

represents the Islamic countries, which consists of 35 countries in total. Similarly, after

some adjustments to missing values in the data, these 35 countries will be incorporated

to several regressions. Therefore, the latest 35 countries are adjusted to a set of 32 and

23 countries as listed in Appendix 9.4 and 9.5 respectively. The variables regressed for

the different samples are summarized by the � symbol in table 8.

Variables Panel-32 Panel-23

Developing � �

Islamic Finance � �

Investment � �

Oil Volatility � �

Inflation Volatility � �

Trade � �

Interest � �

Private credit � �

Table 8- Sub-division of sub-panel regression

Comparing the volatility in 4.1 and 4.2 dataset, we are depicting the graph of the GDP

volatility over time in the below figures. We are only comparing those samples that have

no missing values on any variables as a whole. This justifies the comparison between the

89 countries sampled from regression 4.1 and 23 countries from regression 4.2. The

comparison graphs suggests that the concentration of most countries’ scatter plot lie

around 0-10% on Figure 6 and 0-4% on Figure 7. Therefore, the graphs show a rough

Figure 6- Volatility of 89 countries over

time

Figure 7- Volatility of 23 countries over

time

Student no: 620029286 27

indication of the hypothesis of low GDP volatility of Islamic countries represented by

figure 7 below.

4.3 Method of Estimation

The Panel data proposed has been justified in 4.1 due to its advantages of dealing

successfully with both cross sectional and time series data. However, another aspect

that is important is the method of estimation of the regression tested. There are four

methods as summarized by Verbeek (2004) and adapted to the regression 4.1, in brief it

estimates in the following manner:

1) Between Estimator:

Using Regression in 4.1:

Then compute the individual’s country average across time and total average across

both country and time:

The regression under this method is transformed by subtracting (2) from (1), the

resultant regression:

Then, the coefficients are determined from this new regression. There is a useful

implication to note that under this regression, the oil volatility term is eliminated due

to the nature of the estimator of using the cross sectional part of the data. The oil

volatility term does not change across countries, which justifies its redundancy.

Moreover, the independent variables in this case are exogenous and do not correlate

with the time invariant effects, i.e., the two dummy variables.

2) Within Estimator [Fixed Effects (FE) Estimator]:

Using Regression in 4.1:

Then compute the individual’s country average across time:

Student no: 620029286 28

The regression under this method is transformed by subtracting (4) from (3), the

resultant regression:

Then, the coefficients are determined from this new regression.

Interestingly, the implication in this regression is the mirror image of the method in

regression 1). In this, the time-invariant variables such as Islamic Finance and

Developing Country dummy are therefore eliminated. This is because in contrast to the

method in 1), this method estimates the time series portion of the dataset within a

country individually.

In this case, the independent variables are suppose to be exogenous as well but in

contrast to the method in 1), they are allowed to correlate with the dummy variables.

3) Ordinary Least Squares (OLS) Estimator:

This method exploits both dimension of time and cross section. It is widely known as

the pooled OLS estimator. It directly estimates the original equation in (1) without any

transformation. This is based on the assumption that each observation comes from an

independent, identical distribution (iid), that is . The drawback using this

method is that the panel structure is ignored due to the characteristic of a pooled OLS

method that weighs each observation uniformly. This may not be optimal in practical

data sets. Although the estimates may be unbiased and consistent, it fails to be the

Best Linear Unbiased Estimator (BLUE). The ‘Best’ means the most efficient

estimator, which is, achieving the least variance. Therefore, under panel data, OLS

method might not accomplish that objective of being the ‘Best’.

4) Random Effects Estimator (RE):

The last method in this case is a randomization of between and within estimator. This

method is also known as the Generalized Least Squares (GLS) estimator. Although this

method is a mix between 1) and 2), the strict exogeneity and no correlation are

assumed in this method. The key substance in this method is the determination of .

The value of will determine the extent of randomization of the two extreme

estimators discussed in 1) and 2). The regression in this case is as follows:

θ - √σ2ε / Tσ2ν σ2ε) (5)

Student no: 620029286 29

If θ is 0, the random effects estimator will be identical to the one estimated by OLS

method in 3). If θ is 1, the estimator will then be equal to the one in method 2), that is,

the within estimator.

Since the value of θ is vital, the computation of the entire five samples are presented

below:

Stata computes the value according to equation (5); by running the random effects

regression on the sample with ‘theta’ command.

The usefulness of θ is quite varied in some ways. Not only it can roughly indicate the

tendency of type of estimator that will be estimated under this method, but it also

measures the degree of biasness of the estimator. The extent of biasness can be

explained in detailed in the subsequent Hausman test in this section.

Now the question arises: which model estimates can we rely on for our prediction? The

first method estimates changes in the variables between the countries over a fixed period

of time. The second method indicates within a country, how the extent of change in the

variables affect the GDP volatility over time. However, these areas are not of interest in

this paper.

The Ordinary Least Squares and Random effects method combines both this method and

takes into account the changes across both time and countries. Illustrating these results

in mind, the pooled OLS and RE method deems to be useful for this research.

Hausman test

Hausman (1978) suggested a test to decide in a more convincing manner on the

appropriate method for panel regressions comparing FE and RE. The test is clearly

expounded in Verbeek (2004) where the hypotheses are:

H0: There is no significant correlation between the two dummy variables with the other

independent variables (investment, oil volatility, inflation volatility, trade openness, real

interest rate and credit).

Implication of the null hypothesis: Fixed effects estimator is consistent and not efficient.

Random effects estimator is both consistent and efficient.

Table 9- Computation of transformation parameter in Random Effects Model

Panel-89 Panel-116 Panel-102 Panel-32 Panel-23

θ-Value 0.6381 0.6366 0.7643 0.6806 0.6228

Implication The value is between 0 and 1, more tendencies towards within estimator effect.

Student no: 620029286 30

H1: the null hypothesis of no significant correlation does not hold.

Implication of the alternative hypothesis: Fixed effects estimator is both consistent and

efficient. Random effects estimator is now inconsistent but still efficient.

The distribution is assumed to follow Chi-square asymptotically with degrees of freedom

equal to the dimensions of the estimates.

From both hypotheses, we can deduce that rejection of the null indicates that fixed effects

method is a better explanatory model for the regression tested and the reverse is true on

the failure of rejection of the null.

The test results are as follows for both 4.1 and 4.2 regressions:

Panel-89 Panel-116 Panel-102 Panel-32 Panel-23

Test

statistic 7.19 16.48 5.69 2.24 3.03

P-value 0.3036 0.0024 0.2233 0.8150 0.8055

Decision Reject H0 Fails to reject H0 Reject H0

Implication Choose RE Choose FE Choose RE

Table 10- Hausman test

With the results in the table, we would then proceed with the method suggested by the

tests. Therefore, except for 116 sets of countries in Appendix 9.1, the other sample sizes

suggested that random effects would result in consistent estimates. In the panel of 116

countries, there exist a correlation between the dummy variables and other

independent variables. If random effects continued to be estimated for this sample,

then the results would be biased and the extent of biasness can be illustrated using the

value in Table 9.

Student no: 620029286 31

5. Results

From the previous subsection, we can conclude that focus can be concentrated on two

methods of estimation, which is OLS and RE method. However, for the panel of 116

countries, we would then focus on OLS and FE method justified by Hausman test

results in Table 10.

5.1 Panel Regression

Model

Panel-89 Panel-116 Panel-102

Variables OLS RE OLS FE OLS RE

Developing 0.9924*** (0.1828)

1.204*** (0.3693)

1.4835*** (0.1351)

1.2875*** (0.1610)

1.1570** (0.4746)

Islamic Finance

-0.6079*** (0.1569)

-0.7712** (0.3507)

-0.6299*** (0.1331)

-1.0290*** (0.1678)

-1.1571** (0.5067)

Investment 0.0088* (0.0051)

0.0004 (0.0044)

0.0189*** (0.0038)

0.0033 (0.0032)

0.0017 (0.0012)

-0.0007 (0.0009)

Oil Volatility

-0.4313*** (0.0727)

-0.3749*** (0.0644)

-0.5248*** (0.0622)

-0.4882*** (0.0555)

-0.4015*** (0.0774)

-0.4563*** (0.0606)

Inflation Volatility

0.0022*** (0.0003)

0.0021*** (0.0003)

0.0005*** (0.0001)

0.0005*** (0.0001)

Trade 0.0048*** (0.0013)

0.0044* (0.0025)

0.0045*** (0.0012)

0.0049 (0.0038)

0.0049*** (0.0014)

0.0062** (0.0031)

Interest -0.0119** (0.0061)

-0.0089 (0.0060)

-0.0158*** (0.0056)

-0.0104** (0.0051)

Private credit

-0.0058*** (0.0021)

-0.0041 (0.0032)

Constant 2.3530*** (0.3873)

2.1778*** (0.7807)

1.4212*** (0.3063)

3.0871*** (0.2988)

2.4082*** (0.3757)

2.8641*** (1.0386)

R2 Overall 0.1474 0.1438 0.1247 0.0469 0.0732 0.0686

Table 11- Panel Regression results

Standard errors are shown in parentheses.

* significant at 10% level

**significant at 5% level

***significant at 1% level

Let us consider the first OLS regression on the set of 89 countries, which was run over

all explanatory variables considered under the study. Being a developing country, the

GDP volatility increases by 99.2% more than the developed economies for every unit

change in the volatility. In the same manner, being a country that practices Islamic

Finance, for every change in the volatility, it will reduce by 60.8% more than those

countries that do not practice.

Student no: 620029286 32

A percentage extra invested will affect the GDP volatility in a positive manner by

increasing it by 0.8%. However, every unit increase in oil volatility will affect the GDP

volatility in a negative manner by 43.1%, which can be backed by literature as in Guo

and Kliesen (2005). They stated that when oil prices are too volatile, there is a

reduction in GDP changes, since there is a shift of focus due to uncertainty in resource

allocations. Therefore, investment for growth of the country is delayed, by minimizing

too much change in GDP’s composition.

Moreover, the opposite is seen regarding the effect of inflation volatility. If inflation

volatility increases by one unit, then the GDP becomes more volatile by 0.2%. Following

on, a 1% increase trade, which represents the flows of exports and imports, would

increase the GDP volatility by roughly 0.5%.

There are both negative impacts on the other two variables. Units increase in both real

interest rate and private credit will reduce the GDP volatility by 1.2% and 0.6%

respectively. In general for Fixed effects and Random effects, the direction of changes is

very similar only different in the percentage changes unless some of the estimates are

insignificant. This generalization applies to the different sample size of 116 and 102

countries shown in appendix 9.1 and 9.2 respectively.

Student no: 620029286 33

5.2 Sub-panel Regression

This regression is run on the equation discussed in section 4.2. The main difference

between both panels is the variable Islamic concentration and the countries regressed.

Those considered in this panel are the ones who practice Islamic Banking systems. Due

to limited data for some countries, complete regression was run on 23 countries and 32

countries are regressed without the variable interest rate. The lists of countries are

presented in Appendix 9.5 and 9.4 respectively. The results are presented as follows:

Model

Panel-32 Panel-23

Variables OLS RE OLS RE

Developing 0.4305

(0.3167) 0.8708

(0.7016) 0.7129** (0.3015)

0.9022 (0.5928)

Islamic Concentration 5.0346

(3.1747) 3.4013

(7.9787) -3.2665 (4.3973)

-5.9415 (9.4905)

Investment -0.0042 (0.0067)

-0.0156*** (0.0057)

-0.0240** (0.0094)

-0.0249*** (0.0081)

Oil Volatility -0.2286** (0.0918)

-0.2540*** (0.0773)

-0.1929** (0.0978)

-0.1823** (0.0853)

Inflation Volatility 0.0066*** (0.0008)

0.0047*** (0.0007)

0.1222*** (0.0174)

0.0925*** (0.0159)

Trade 0.0066*** (0.0016)

0.0113*** (0.0033)

0.0073*** (0.0014)

0.0109*** (0.0029)

Interest

-0.0262* (0.0142)

-0.0046 (0.0140)

Private credit -0.0064** (0.0030)

-0.0047 (0.0053)

0.0028 (0.0030)

0.0031 (0.0050)

Constant 1.7324** (0.6746)

0.6951 (1.4369)

0.4844 (0.6519)

-0.1658 (1.2156)

R2 Overall 0.2034 0.1706 0.2199 0.1910

Table 12- Sub-panel Regression results

Standard errors are shown in parentheses.

* significant at 10% level

**significant at 5% level

***significant at 1% level

The sub-panel regression results that is significant here have similar direction in terms

of their movements in relation to GDP volatility, except for Investment. The movement

of this variable is an interesting finding since now it is a negative relation contrast to

5.1 where it was positive. This is then now interpreted that taking into account

investments in countries that practice Islamic Finance only leads to less volatile GDP

Student no: 620029286 34

movement by at least 1.61% in contrast to the investments under a general economy in

5.1 regression increases the volatility.

On the other hand, regarding the purpose of the sub-panel regression, that is, the

inclusion of Islamic concentration appears to be insignificant in both sample data. This

can be due to the definition of the variable in this case which may not lead to a good

depiction of the Islamic concentration in the country. Therefore, at a glance, this sub-

panel may appear futile to estimate. Nevertheless, this regression has been able to

prove that investment in the selected Islamic countries induces the GDP to be less

volatile as compared to the 5.1 regressions.

5.3 Panel Validity tests

The two methods of coefficient estimates of Ordinary Least Squares and Random

Effects have been presented in the previous section. However, tests need to be

performed on whether the results of the estimates can be relied upon. This is necessary

as the interpretation of the estimates given above would only be correct if there is an

absence of serial correlation in the residuals and heteroskedasticity of the variance of

the estimates.

McGregor (n.d.) have visually illustrated the tests to perform in order to select the most

appropriate method of estimation. The figure below represents his illustration and

using that, the two tests performed subsequently will be justified.

Student no: 620029286 35

Source: McGregor6 (n.d.)

Figure 8- Flowchart of tests to select appropriate Method of Estimation

Ordinary Least Squares VS Random Effects

This test is a Lagrange Multiplier (LM) test by Breusch and Pagan (1979) testing for

any difference of weightage in individual observation added to sample. This test follows

chi-square distribution asymptotically with 1 degree of freedom. The test hypotheses

are as follows:

H0: There are no individual effects on each observation, that is σ2ν =0 in the equation

(5)

Implication of the null hypothesis: if the variance of individual observation reduces to

zero, a simple OLS estimator is applicable for the regression.

H1: the null hypothesis of no individual effects does not hold.

Implication of the alternative hypothesis: Rejection of the null denotes that there are

some different weightage in some observations in the data. Random estimator is

therefore appropriate.

6 Neil Foster McGregor presented the flowchart in his lecture slides of Lecture 6-Panel Data Econometrics at

the University of Vienna.

Student no: 620029286 36

Model

Test Panel-89 Panel-116 Panel-102 Panel-32 Panel-23

OLS VS RE 672.40

(0.0000) N.A.

1218.57 (0.0000)

390.46 (0.0000)

193.14 (0.0000)

P-values are presented in the parentheses.

Table 13- Model Selection test A

The test for all the samples therefore clearly rejects the null indicating that random

effects is a better model, substantiated by the p-value in the parentheses that are below

5% level.

Ordinary Least Squares VS Fixed Effects

With the help of Gregory Chow of Princeton University’s paper in 1960, we can

continue to select the most appropriate model by running a Chow test.

This test has the following hypotheses:

H0: There are no significant differences on each panel effect, that is, the individual

country effects are jointly zero.

Implication of the null hypothesis: if the effect of joint country effect reduces to zero, a

simple OLS estimator is applicable for the regression.

H1: the null hypothesis of no significant difference does not hold.

Implication of the alternative hypothesis: Rejection of the null denotes that there are

some different weightage in some country effect in the data. Within estimator (Fixed

Effects) is therefore appropriate.

It is therefore computed as follows:

Where:

RRSS- Restricted Residual Sum of Squares (if null hypothesis is true)

URSS- Unrestricted Residual Sum of Squares (if null hypothesis is false)

N-number of panel (countries)

T- number of period (years)

K- number of parameters tested.

Student no: 620029286 37

In this particular case, The Degrees of freedom is as follows:

Numerator: N-1= 116-2-1 = 113 (two of the countries eliminated due to multi-

collinearity.

Denominator: NT-N-K= {(116x17)- 116-119} = 1737

The critical value of F-distribution at (113, 1737) computed using excel formula at 5%

significance level = 1.2382.

Model

Test Panel-89 Panel-116 Panel-102 Panel-32 Panel-23

OLS VS FE N.A. 9.43

(0.0000) N.A. N.A. N.A.

P-value is presented in the parentheses.

Table 14- Model Selection test B

Here, the clear rejection is not only supported by the p-value below 5%, but also by the

test statistic of 9.43> 1.2382 (critical value at 5% significance level). This concludes

that we should use the Fixed Effects method to estimate the coefficients in this model.

Serial correlation

The presence of serial correlation in the residuals will cause the standard errors to be

smaller than the actual values. It is therefore important to detect if such phenomenon

exists, and then correct them. This test is conducted by using ‘xtserial’ command on the

earlier regressions in 5.1. Stata uses Woolridge test for serial correlation. Drukker

(2003) supports the Woolridge test due to its easy implementation and applications. In

the paper, he also simulated some evidence that the test has good size and power

properties in its application, making it more robust to rely on. This method does that by

using the residuals from the first-difference regressions of the variables. The

hypotheses tested are as follows:

H0: There is no serial correlation in the residuals, that is condition holds.

Implication of the null hypothesis: The correlation between the residuals will enable us

to rely on the standard errors estimated in the regression tested.

H1: the null hypothesis of no serial correlation between the error terms does not hold.

Implication of the alternative hypothesis: Rejection of the null denotes the reliance on

the standard errors in the regression is specious.

Student no: 620029286 38

Model

Test Panel-89 Panel-116 Panel-102 Panel-32 Panel-23

Serial Correlation

126.39

(0.0000) 112.56

(0.0000) 151.70

(0.0000) 187.28

(0.0000) 60.20 (0.0000)

P-value is presented in the parentheses.

Table 15- Serial Correlation validity tests

All the sample data rejects the null at 5% level, strong support of presence of serial

correlation in the error terms.

Heteroskedasticity

The presence of heteroskedasticity in the variance of the residuals breaks the

assumption of constant variance, therefore reduces the predictability of the estimates.

Consequently, it is important to verify that homoskedasticity assumption still holds.

There are two ways to conduct the test. The first test is executed by using the command

‘xttest3’ on the fixed effects regression. Stata by default uses a modified Wald Test to

perform the group-wise heteroskedasticity tests as specified in Torres (n.d.). According

to Verbeek (2004), the Heteroskedasticity test used for fixed effects regression can also

be used for Random Effects as long as the time invariant variable is independent and

identically distributed (i.i.d.). We have tested that the variable is iid as long as Random

effects is the model appropriate for the regression. This was confirmed in section 4.3

under - Hausman test.

The second is Poi and Wiggins (2001) Likelihood Ratio test. As explained by Sanchez

(2012), the test is executed by running ‘xtgls’ command on the regression of interest.

This method uses Maximum Likelihood to model the panel level heteroskedasticity.

Then, the following is to restrict homoscedastic assumption to the error terms. Using

the likelihood values has the benefit of looking whether the additional parameters

(covariances between panels) improves the model fit. The test is calculated using the

formulae:

Where denotes the log-likelihood function of the unrestricted and the log

likelihood function of the restricted estimator. The test follows a Chi-square

distribution based on asymptotically normal distribution.

Here, the first method is used and the results are as follows:

Student no: 620029286 39

Model

Test Panel-89 Panel-116 Panel-102 Panel-32 Panel-23

Heteroskedasticity-Wald test

160000 (0.0000)

90954.26 (0.0000)

120000 (0.0000)

12509.47 (0.0000)

7053.85 (0.0000)

P-value is presented in the parentheses.

Table 16- Heteroskedasticity validity test

From the Wald test, all the dataset leads to rejection of the null hypothesis, indicating

the presence of heteroskedasticity.

In the light of the results of all the above tests, the coefficients in Section 5 cannot be

relied upon due to its heteroskedastic and serially correlated error terms.

Therefore, we need to transform the regression results to take into account the

heteroskedasticity and serial correlation present on the panels.

In Table 17 below, the coefficient estimates are the same as presented in the results in

Section 5. The main modification is that the standard error in the parentheses are

estimated again to be robust to any heteroskedastic and serial correlation problem that

might affect the reliance on the estimates. Another important finding to note is that

some of the significant results earlier have deemed to be insignificant due to

transformed standard errors. This includes the investment in the sample of 32

countries, inflation volatility in 116 countries and the trade variable in 89 and 102

countries. However, the variables relate to each other in the same manner, hence the

sign is preserved. The Islamic Finance variable negatively affecting the GDP volatility

remains significant and noteworthy of discussion.

Student no: 620029286 40

Model

Panel-89 Panel-116 Panel-102 Panel-32 Panel-23

Variables RE FE RE RE RE

Developing 1.204*** (0.3619)

1.1570** (0.5252)

0.8708 (0.7901)

0.9022 (0.6978)

Islamic Finance

-0.7712** (0.3428)

-1.1571** (0.3670)

N.A. N.A.

Islamic Concentration

N.A. N.A. N.A. 3.4013

(7.1327) -5.9415

(7.1671)

Investment 0.0004

(0.0088) 0.0033

(0.0081) -0.0007 (0.0005)

-0.0156* (0.0095)

-0.0249 (0.0185)

Oil Volatility -0.3749***

(0.0814) -0.4882***

(0.1063) -0.4563***

(0.0998) -0.2540** (0.1140)

-0.1823* (0.1057)

Inflation Volatility

0.0021** (0.0008)

0.0005 (0.0004)

0.0047* (0.0028)

0.0925 (0.0819)

Trade 0.0044

(0.0031) 0.0049

(0.0068) 0.0062

(0.0044) 0.0113** (0.0045)

0.0109*** (0.0041)

Interest -0.0089 (0.0084)

-0.0104* (0.0061)

-0.0046

(0.0150)

Private credit -0.0041 (0.0043)

-0.0047 (0.0094)

0.0031 (0.0085)

Constant 2.1778*** (0.6785)

3.0871*** (0.5033)

2.8641*** (1.1068)

0.6951 (1.6307)

-0.1658 (1.5387)

R2 Overall 0.1438 0.0469 0.0686 0.1706 0.1910

Table 17- Transformed final model robust to heteroskedasticity and correlation

Student no: 620029286 41

6. Discussion

The present discussion is going to focus on the transformed model in Table 17.

Neumeyer and Perri (2005) has motivated this paper to take into account the difference

in behavior between developed and developing countries. The inclusion of the

Developing dummy variable has been fruitful in explaining that the Developing

country increases the GDP volatility based on the larger sample data regressions.

The creation of the sub-panel was motivated by Beck et al. (2013) who divided the

sample to explain the growth of Islamic Banks. Although the key variable of this paper,

that is, the Islamic Finance dummy variable has successfully explained the lower GDP

volatility effect during the period, the percentage of concentration included in the sub-

panel regressions fails to explain the same.

With this variable, we have tested Chisthi’s (1985) theoretical micro-investment model

on practical data sets. The finding therefore supports Chisthi’s (1985) adaptation of

Minsky Investment Theory that was built up to prove that the amplitude of the

business cycle in an Islamic Economy is lower than in a conventional economy, which

was implemented on a micro-firm level.

Another very important variable included in the model that was motivated by several

papers like Gali and Gambetti (2009), Justiano and Primiceri (2008), Benati and Surico

(2009), Salahuddin and Islam (2008) is investment. The emphasis placed on this

variable has attested to be effective. In the main panel, the effect was positive

indicating that higher investment will increase GDP volatility. Kayed and Hassan

(2011) have found that the conservative nature of investments usually made through

the Islamic system, leads to a more productive outcome, creating stability in the

economic system. In the sub-panel, the investment done in Islamic countries indeed

reduces the GDP volatility, which is in line with the literature by Kayed and Hassan

(2011). This therefore supports the theoretical model by Chisthi (1985) discussed

earlier in this section. The model will therefore be useful for future similar studies in

this area of research.

The importance of oil volatility seemed quite obvious in the results presented in Table

17, which is in line with Imam and Kpodar (2010) viewpoint. Inflation volatility and

GDP volatility move in the same direction independently of the sample difference. The

movement is as predicted in Kent et al. (2005).

Student no: 620029286 42

Trade openness is another structural indicator necessary to explain GDP movements as

highlighted by Kent et al. (2005). Again the direction, which is as, predicted in Kent et

al. (2005) holds for all sample sizes.

After transformation of the model, some of the interest rates coefficients are now

insignificant. Nonetheless in general, the model with random effects of 102 sample

countries proved to be significant and coincides with the literature that interest rate is

acyclical or countercyclical to the business cycle in Neumeyer and Perri (2005).

Following a similar reasoning, the private credit given for loans in countries is not

significant to explain the GDP volatility in contrast to Kent et al. (2005) where there

was a significant positive relation in one of the panel data models that they tested.

Overall, the literature from various authors referred to in the literature review and the

discussion has influenced this research in several areas. The contributions from those

authors have been compiled and adapted to provide a remedy to the research problem

in this paper.

Student no: 620029286 43

7. Limitations and Further Research

There are several inherent limitations that might affect the results of this paper. The

data for Islamic banking concentration is found by number of banks registered and set

up in each country and not taking into account the total number of branches of those

registered banks. The access to the number of branches compared to conventional

banks would be a better variable in the sub panel regression. This data would provide a

better representation of the Islamic Banking presence in each country, as it would

consider the intensity of the use of this banking system.

Moreover, the listed Islamic Banks in itself may not have administered according to the

binding principles stated in Quran and Sunnah, also popularly known as Sharia law.

According to Chong and Liu (2009) their banking operations try to imitate the

conventional system in order to remain competitive in the countries they are

established. As Beck et al. (2013) also supported this in their paper since they found

that the products offered in the Conventional and Islamic systems are structured in a

similar way and ‘a strong element of equity participation’ (p. 438) is found in the latter

system. Therefore, there is lack of uniformity in the implementation of the banking

principles, which hinders comparison of the data collected. As a result, this issue

imposes some inference limit on the generalization of the results presented in Table 17.

Further innovation can be adapted from Kent et al. (2005) discussion where they

suggested using the United States (US) data as a control country to compute the new

variables. Since their study did not include the Islamic Finance variable, further

research could consider using US as a control along with the Islamic Finance variable

to explain Great Moderation. The motivation of using United States as control would be

justified by the concept of Great Moderation being the country of its establishment.

The regression would therefore be:

The ‘Tilda’ as explained by Kent et al. (2005) represents the respective country’s

observation difference with the United States at a fixed period of time for each

observation. The oil volatility therefore will not change due to its property of being

capped to US oil price, which is universally the same at a fixed period of time.

Student no: 620029286 44

Another way of observing the hypotheses tested is by using time series approach. If

time series data were to be used focusing on United States, variations on the tests

could be done. Granger causality effects could be tested on the growth of Islamic

finance throughout the years affecting the GDP volatility. If the Islamic Finance

growth over the years appears to granger cause the GDP volatility, then it means the

former variable will be able to provide statistical significant information about the

latter variable. However, this effect does not mean the Islamic Finance variable causes

the changes in the volatility. It only aids in contributing significant information in

prediction.

Student no: 620029286 45

8. Conclusion

The changes in GDP volatility have been universally accepted to explain business cycle

fluctuations according to umpteen research papers. The Great Moderation concept was

quite a popular research motivation during its reigning years until the financial crisis

around the year 2007 took over its place. Since then not many papers continued to

pursue the understanding of this Great concept which was defined as a period of low

volatility in the business cycle over 1985-2007. According to the literature review, we

would classify it as a grey area since there is no consensus in the reasons of its

existence.

Our intuition is that the consensus not found in the literature might be due to some

missing explanatory variables such as Islamic Finance. The innovation introduced by

this paper was considering the presence of Islamic Finance as a determinant of GDP

volatility adding to the literature on this subject of interest. We found that indeed the

presence of Islamic Finance helps to reduce GDP volatility. This could be one of the

possible reasoning behind Great Moderation occurrence. The empirical regressions

provided support to the hypothesis tested and contributed to a more profound

understanding of the role of Islamic Finance in explaining the business cycle. Not only

would this finding be useful for understanding the past, but also it would help to

forecast and aid in policy making for the future.

Student no: 620029286 46

9. Appendices

9.1 Panel 116 countries

Albania Cote d'Ivoire Korea, Rep. Portugal

Algeria Cyprus Kyrgyz Republic Romania

Argentina Czech Republic Latvia Russian Federation

Armenia Denmark Lesotho Rwanda

Australia Dominican Republic Lithuania Senegal

Austria Ecuador Luxembourg Singapore

Azerbaijan Egypt, Arab Rep. Macao SAR, China Slovak Republic

Bahamas, The El Salvador Macedonia, FYR Slovenia

Bangladesh Estonia Madagascar South Africa

Barbados Ethiopia Malaysia Spain

Belarus Finland Mali Sri Lanka

Belgium France Malta Sudan

Belize Gabon Mauritania Swaziland

Benin Gambia, The Mauritius Sweden

Bolivia Germany Mexico Switzerland

Botswana Greece Moldova Syrian Arab Republic

Brazil Guatemala Morocco Tanzania

Brunei Darussalam Hong Kong SAR, China Mozambique Thailand

Bulgaria Hungary Myanmar Togo

Burkina Faso Iceland Netherlands Trinidad and Tobago

Cambodia India New Zealand Tunisia

Cameroon Indonesia Norway Turkey

Canada Iran, Islamic Rep. Pakistan Uganda

Cape Verde Israel Panama Ukraine

Chad Italy Papua New Guinea United Kingdom

China Japan Paraguay United States

Congo, Dem. Rep. Jordan Peru Uruguay

Congo, Rep. Kazakhstan Philippines Yemen, Rep.

Costa Rica Kenya Poland Zambia

Student no: 620029286 47

9.2 Panel-102 countries

Albania Finland Nicaragua

Algeria France Norway

Argentina Gabon Panama

Armenia Gambia, The Papua New Guinea

Australia Germany Paraguay

Bahamas, The Greece Peru

Bangladesh Guatemala Philippines

Barbados Hong Kong SAR, China Poland

Belarus Hungary Romania

Belgium Iceland Russian Federation

Belize India Rwanda

Bolivia Indonesia Sierra Leone

Botswana Israel Singapore

Brazil Italy Slovak Republic

Brunei Darussalam Japan Slovenia

Bulgaria Jordan South Africa

Cameroon Kenya Spain

Canada Korea, Rep. Sri Lanka

Cape Verde Latvia Swaziland

Chad Lesotho Sweden

Chile Lithuania Switzerland

China Macao SAR, China Syrian Arab Republic

Congo, Rep. Macedonia, FYR Tanzania

Costa Rica Madagascar Thailand

Croatia Malaysia Trinidad and Tobago

Cyprus Malta Uganda

Czech Republic Mauritania Ukraine

Denmark Mauritius United Kingdom

Dominican Republic Mexico United States

Ecuador Moldova Uruguay

Egypt, Arab Rep. Morocco Venezuela, RB

Equatorial Guinea Myanmar Vietnam

Estonia Namibia Yemen, Rep.

Ethiopia Netherlands Zambia

Student no: 620029286 48

9.3 Panel-89 countries

Albania Gabon Panama

Algeria Gambia, The Papua New Guinea

Argentina Germany Paraguay

Armenia Greece Peru

Australia Guatemala Philippines

Bahamas, The Hong Kong SAR, China Poland

Bangladesh Hungary Romania

Barbados Iceland Russian Federation

Belarus India Rwanda

Belgium Indonesia Singapore

Belize Israel Slovak Republic

Bolivia Italy Slovenia

Botswana Japan South Africa

Bulgaria Jordan Spain

Cameroon Kenya Sri Lanka

Canada Korea, Rep. Swaziland

Cape Verde Latvia Sweden

Chad Lesotho Switzerland

China Lithuania Syrian Arab Republic

Congo, Rep. Macao SAR, China Tanzania

Costa Rica Macedonia, FYR Thailand

Cyprus Madagascar Trinidad and Tobago

Czech Republic Malaysia Uganda

Denmark Malta Ukraine

Dominican Republic Mauritius United Kingdom

Ecuador Mexico United States

Egypt, Arab Rep. Morocco Uruguay

Ethiopia Myanmar Yemen, Rep.

Finland Netherlands Zambia

France Norway

9.4 Panel-32 countries

Algeria Azerbaijan Bangladesh Cyprus Egypt, Arab Rep. Ethiopia Gambia, The Indonesia Iran, Islamic Rep. Israel Jordan

Kazakhstan Kenya Malaysia Mauritius Pakistan Philippines Russian Federation Senegal Singapore South Africa Sri Lanka

Sudan Switzerland Syrian Arab Republic Tanzania Thailand Tunisia Turkey Uganda United Kingdom United States

Student no: 620029286 49

9.5 Panel-23 countries

Algeria Bangladesh Cyprus Egypt, Arab Rep. Ethiopia Gambia, The Indonesia Israel Jordan Kenya Malaysia Mauritius

Philippines Singapore South Africa Sri Lanka Switzerland Syrian Arab Republic Tanzania Thailand Uganda United Kingdom

United States

Student no: 620029286 50

10. Glossary

Gharar: an element of Uncertainty

Maysir: an element of Speculation

Musharakah: A joint partnership contract with profit/loss sharing agreements instead

of interest bearing loan contracts.

Quran: The Islam Recitation.

Riba: unjust gains in trade or business: usually synonymous with interest

Sharia: The rules and underlying principles of Islamic law.

Sunnah: Usual practices of Islam that was taught by Prophet Muhammad (PBUH)

Student no: 620029286 51

11. References

Abu Dhabi Islamic Bank, n.d. Understanding Islamic Banking. Available from:

http://www.adib.ae/understanding-islamic-banking-0 (Accessed 01 July 2013).

Association of Islamic Banking Institutions Malaysia, n.d. Islamic Financial

Institutions. Available from:

http://aibim.com/component/option,com_bookmarks/Itemid,99999999/mode,0/catid,5/na

vstart,1/search,search../ (Accessed 15 July 2013).

Baltagi, B.H., 2005. Econometric Analysis of Panel Data. 3rd Edition. Chichester: John

Wiley & Sons Ltd.

Bashir, A.H.M.,2007. Islamic Banks Participation, Concentration and Profitability:

Evidence From MENA Countries. Economic Research Forum. WP/0402.

Beck, T., Kunt, A.D. and Merrouche, O., 2013. Islamic Vs Conventional Banking:

Business Model, Efficiency and Stability. Journal of Banking and Finance, 37(2), 433-

447.

Benati, L. and Surico, P., 2009. VAR Analysis and the Great Moderation. American

Economic Review, 99(4), 1636-1652.

Burns, A.F. and Mitchell, W.C., 1946. Measuring Business Cycles. National Bureau of

Economic Research.

Chishti, S. U., 1985. Relative Stability of Interest Free Economy. Journal of Research

in Islamic Economics, 3(1), 3-12.

Chong, B.S. and Liu, M.H. 2009. Islamic Banking: Interest-Free or Interest-Based?

Pacific-Basin Finance Journal, 17(1), 125–144.

Student no: 620029286 52

Creal, D., Koopman, S.J. and Zivot, E., 2010. Extracting A Robust US Business Cycle

Using A Time-Varying Multivariate Model-Based Bandpass Filter. Journal of Applied

Econometrics, 25(4), 695-719.

Drukker, D.M., 2003. Testing for serial correlation in linear panel-data models. The

Stata Journal, 3(2), 168-177.

Easterly, W., Islam, R. and Stiglitz, J.E., 2000. Shaken and Stirred: Explaining Growth

Volatility. Annual Bank Conference on Development Economics. Washington, DC:

World Bank, 191-212.

Gali, J. and Gambetti, L., 2009. On the Sources of the Great Moderation. American

Economic Journal: Macroeconomics, 1(1), 26-57.

Guo, H. and Kliesen, K.L., 2005. Oil Price Volatility and US Macroeconomic Activity.

Federal Reserve Bank of St. Louis Review, 87(6), 669-684.

Imam, P. and Kpodar, K., 2010. Islamic Banking: How Has it Spread? IMF Working

Paper, WP/10/195.

International Monetary Fund (IMF), 2012. Global Financial Stability Report: Restoring

Confidence and Progressing on Reforms. Washington, DC: IMF.

Johnston, J. and DiNardo, J., 1997. Econometric Methods. 4th Edition. New York:

McGraw-Hill.

Justiniano, A. and Primiceri, G.E., 2008. The Time-Varying Volatility of

Macroeconomic Fluctuations. American Economic Review, 98(3), 604-641.

Student no: 620029286 53

Kayed, R.N. and Hassan, M.K., 2011. The Global Financial Crisis and Islamic Finance.

Thunderbird International Business Review, 53(5), 551-564.

Kent, C., Smith, K. and Holloway, J., 2005. Declining output volatility: what role for

structural change? In: Kent, C. and Norman D. RBA Annual Conference. The Changing

Nature of the Business Cycle Reserve Bank of Australia.

McGregor, N.F., n.d. Lecture 6 Panel Data Econometrics. Available from:

http://homepage.univie.ac.at/Neil.Foster/TEACHING/ECONO/Lec06%20-

%20Panel%20Data.pdf (Accessed 25 July 2013)

Murdoch, G., 2009. TIMELINE: Milestones in rise of Islamic Finance. United States:

Reuters. Available from:

http://www.reuters.com/article/2009/04/13/us-islamicbanking-summit-timeline-

idUSTRE53C12520090413 (Accessed 15 July 2013)

Neumeyer, P.A. and Perri F., 2005. Business Cycles in Emerging Economies: The Role

of Interest Rates. Journal of Monetary Economics, 52(2), 345-380.

Salahuddin, M. and Islam, Md.R., 2008. Factors Affecting Investment In Developing

Countries: A Panel Data Study. The Journal of Developing Areas, 42(1), 21-37.

Sanchez, G., 2012. Fitting Panel Data Linear Models in Stata. StataCorp LP.

Sole, J., 2007. Introducing Islamic Banks into Conventional Banking Systems. IMF

Working Paper. WP/07/175.

Torres, O.R.,n.d. Panel Data Analysis Fixed and Random Effects (using Stata 10.x).

Version 4.1. Available from:

http://www.princeton.edu/~otorres/Panel101.pdf (Accessed 20 July 2013)

Student no: 620029286 54

The International Statistical Institute, 2013. Developing Countries. Available From:

http://www.isi-web.org/component/content/article/5-root/root/81-developing (Accessed

20 July 2013).

Verbeek, M., 2004. A guide to Modern Econometrics. 2nd Edition. Chichester: John

Wiley & Sons Ltd.

Student no: 620029286 55