The Effect of Foreign Direct Investments on the Level of Output and its Economic Growth;
Investigating the Solow growth Model and the issue of Convergence, Does Grouping Matter?
by
Kolthoom Alkofahi, B.S., M.A.
A Dissertation
In
Economics
Submitted to the Graduate Faculty
of Texas Tech University in
Partial Fulfillment of
the Requirements for
the Degree of
DOCTOR OF PHILOSOPHY
Approved
Dr. Masha Rahnamamoghadam
Chair of Committee
Dr. Terry Von Ende
Dr. Robert P. Mccomb
Mark Sheridan
Dean of the Graduate School
May, 2014
Copyright 2014, Kolthoom Alkofahi
Texas Tech University, Kolthoom Alkofahi, May 2014
ACKNOWLEDGMENTS
It’s been a long road, but here I am at the end, where there are so many people to who
thanks I extend!
First and foremost, I would like to thank “Allah” for guiding and giving me the strength
and the ability to get through this amazing learning experience, and granting me the chance to
study and work with great people at Economics department. The lovely and friendly environment
encouraged me to work at my best, for all of you there I say “thank you very much”.
I would also like to express my deep and sincere gratitude to my advisor Professor Masha
Rahnama for his guidance, constructive comments and support through this work. Without his
guidance and persistent help this dissertation would not have been possible.
A thank you should be devoted to my committee members, Professor Terry Von Ende
and Professor Robert McComb, who always listened and supported me whenever I felt down.
Your words of wisdom have inspired me to never give up and continue to face the problems as a
young Economist until I reach my goal.
In addition, a thank you to Professor Peter Summers who introduced me to some of the
software packages and gave me some hints on how to deal with a special problems that
encountered.
My husband, Dr. Hisham Bani-Salameh, you have supported me emotionally and
financially, you were always there whenever I felt discouraged by the ups and downs through my
work. You stood as a wise man at one hand, and at the other, your endless love make the whole
wide world as shiny as your lovely smile. I can’t find enough words to express my thanks to you,
for you I say “I can’t stop loving you”.
Never forget the role that my Kids played in cheering me up. Layth, Layan, and Karam, I
truly can’t live without you because you are such the beautiful world that I only can live in,
otherwise, my life is just useless. I love you so much, and wish that I would be a role model for
you to build up your career. I also wish that the sacrifices and help that you all made give you
some idea on how hard and diligent you should work to reach your goal in this life.
ii
Texas Tech University, Kolthoom Alkofahi, May 2014
Mom, Amineh Alkofahi, you are the reason why I am at this stage. Your endless love and
support at each level of my life, the way you built my personality are the paradigm that I always
follow. My dad, Abdolaziz Alkofahi, all the words that I know will never give you what you
deserve; you are the sweetest, lovely, and giving father in the whole wide world. My mom and
dad, I love you so much and will always ask Allah to protect you and grant all your wishes.
My eight siblings, Amal, Natheer, Raief, Umaiah, Hussam, Hisham, Heba, Khawla, there
are so many things I would like to say about you, however, at this stage, I would like to thank
you all for the love and supports that you always giving me, you believed in me to grant my
Mom’s wish, you always consider me the lovely younger sister, and always encourage me to
reach where I am standing right now. I thank Allah from my heart for such lovely family that I
am surrounded by. Thank you so much and love you all the time.
I would like to thank all my family in Law, my friends, neighbors, and everyone who had
confidence in me. Thank you so much and may Allah grant all your wishes and give you a good
life and a good health.
iii
Texas Tech University, Kolthoom Alkofahi, May 2014
Table of Contents
ABSTRACT............................................................................................................................ vii
LIST OF TABLES ............................................................................................................... viii
LIST OF FIGURES ................................................................................................................. x
INTRODUCTION ..................................................................................................................... 1
LITERATURE REVIEW ......................................................................................................... 8
ECONOMIC GROWTH AND THE ISSUE OF CONVERGENCE ................................................................. 8
FDI AND HUMAN CAPITAL ACCUMULATION ................................................................................... 15
FDI, ECONOMIC GROWTH, AND CONDITIONAL CONVERGENCE ...................................................... 16
Microeconomic prospective .................................................................................................... 17
Macroeconomics prospective ................................................................................................. 18
DATA AND SAMPLES .......................................................................................................... 22
DATA ............................................................................................................................................. 22
SAMPLES ........................................................................................................................................ 24
AN OVERVIEW OF FOREIGN DIRECT INVESTMENT AND ITS GLOBAL TREND . 27
DIFINETIONS OF FOREIGN DIRECT INVESTMET ............................................................................... 27
TYPES OF FDI .................................................................................................................................. 29
MOTIVATIONS FOR FDI .................................................................................................................. 30
FDI and SPILLOVERS ....................................................................................................................... 30
FACTORS THAT HELP BROADEN FDI ................................................................................................ 31
GLOBAL FDI TRENDS....................................................................................................................... 31
METHODOLOGY, EMPIRICAL MODELS, AND DISCUSSION OF RESULT .............. 35
A. TEXTBOOK SOLOW GROWTH MODEL ........................................................................................ 36
FIRST APPROACH: OLS CROSS COUNRY FRAMEWORK .................................................................... 44
DISCUSSION OF RESULTS ................................................................................................................ 46
CASE I: SAMPLES OF MRW ...................................................................................................... 46
CASE II: DEVELOPING COUNTRIES AND SUBSAMPLES .............................................................. 49
B. THE AUGMENTED SOLOW MODEL ............................................................................................ 52
iv
Texas Tech University, Kolthoom Alkofahi, May 2014
DISCUSSION OF THE RESULTS ......................................................................................................... 55
CASE I: SAMPLES OF MRW ...................................................................................................... 55
CASE II: DEVELOPING COUNTRIES AND SUBSAMPLES .............................................................. 58
THE ROLE OF FDI AND THE ISSUE OF CONVERGENCE............................................ 61
a) TESTS FOR UNCONDTIONAL CONVERGENCE .............................................................................. 65
DISCUSSION OF RESULTS ................................................................................................................ 65
CASE I: SAMPLES OF MRW ...................................................................................................... 65
CASE II: DEVELOPING COUNTRIES AND SUBSAMPLES .............................................................. 68
b) TESTS FOR CONDTIONAL CONVERGENCE .................................................................................. 70
1. THE TEXTBOOK (BASIC) SOLOW MODEL .............................................................................. 70
DISCUSSION OF RESULTS ................................................................................................................ 70
CASE I: SAMPLES OF MRW ...................................................................................................... 70
CASE II: DEVELOPING COUNTRIES AND SUBSAMPLES .............................................................. 73
2. THE CONDITIONAL CONVERGENCE BASED ON FDI ..................................................................... 76
DISCUSSION OF RESULTS ................................................................................................................ 77
CASE I: SAMPLES OF MRW ...................................................................................................... 77
CASE II: DEVELOPING COUNTRIES AND SUBSAMPLES .............................................................. 81
COMMENT- CROSS SECTIONAL FRAMWORK .................................................................................. 85
PANEL DATA ANALYSIS .................................................................................................... 86
CHOICE OF ESTIMATOR .................................................................................................................. 87
TYPE OF TESTS ............................................................................................................................... 88
ESTIMATION RESULTS .................................................................................................................... 91
Non-Oil Sample:...................................................................................................................... 92
Intermediate Sample .............................................................................................................. 94
The OECD Sample ................................................................................................................... 97
Sample of Developing Countries ........................................................................................... 100
Sample of High Income Developing Countries ....................................................................... 103
Middle Income Developing Sample ....................................................................................... 107
Low Income Developing Countries ........................................................................................ 110
COMMENTS-PANEL DATA ANALYSIS............................................................................................. 113
TESTING FOR CONDITIONAL CONVERGENCE PREDICTIONS OF THE SOLOW
AND AUGMENTED SOLOW MODELS............................................................................ 115
v
Texas Tech University, Kolthoom Alkofahi, May 2014
DISCUSSION OF RESULTS: SAMPLES OF MRW ............................................................................... 120
CASE I: CONVERGENCE IN SOLOW MODEL ............................................................................ 120
CASE II: CONVERGENCE IN THE AUGMENTED SOLOW MODEL .............................................. 122
COMMENTS: THE ROLE OF FDI ON ECONOMIC GRWOTH USING FIXED EFFECTS APPROACH AND
SAMPLES OF MRW .................................................................................................................. 124
DISCUSSION OF RESULTS: DEVELOPING AND SUBDEVELOPING SAMPLES ..................................... 132
CASE I: CONVERGENCE IN SOLOW MODEL ............................................................................ 132
COMMNTS: THE ROLE OF FDI ON ECONOMIC GRWOTH USING FIXED EFFECTS APPROACH, AND
DEVELOPING AND SUBDEVELOPING SAMPLES......................................................................... 135
STUDY CASE: THE ISSUE OF CONDITIONAL CONVERGENCE IN ISLAM (1995) .......................... 146
FDI AND THE ISSUE OF CONVERGENCE UNDER THE GMM ESTIMATION ........ 151
CROSS-COUNTRY GROWTH EXAMPLES USING GMM ESTIMATORS ............................................... 153
ESTIMATING THE SOLOW GROWTH MODEL ................................................................................. 156
DISCUSSION OF RESULTS BASED ON GMM ESTIMATION .............................................................. 157
COMMENTS ON USING GMM ESTIMATOR ................................................................................... 165
FINAL CONCLUSIONS ...................................................................................................... 174
BIBLIOGRAPHY ................................................................................................................. 177
APPENDIX ............................................................................................................................ 182
vi
Texas Tech University, Kolthoom Alkofahi, May 2014
ABSTRACT
The main objective of this paper is to study the effect of foreign direct investment on the
level of output and its economic growth using recent growth theories and econometric
techniques. Incorporating different groups of countries, we conducted cross sectional framework,
panel data analysis, and dynamic panel estimation in the form of GMM estimation. We started
the study by taking the work of Mankiw, Romer, and Weil (MRW, 1992) to test the validity of
the Solow model in explaining income differences across countries. We constructed more
comprehensive, revised, and extended data that covers the time period of 1980 to2010 and
included better constructed groups of countries. We augmented the model with foreign direct
investment and tested if it can further improve the results, and if it can be considered as a factor
that helps explain income differences across countries.
The issues of unconditional and conditional convergence are also considered in the
study, with and without incorporating FDI. The results did not support the cross sectional
framework for samples in MRW, nor the samples we constructed. Foreign Direct Investment is
found to be positive, significant, growth enhancing engine, and important factor in explaining
income differences across countries when panel estimation is employed. To form a
comprehensive analysis, two types of GMM estimators were employed. The results provide more
evidence of conditional convergence. The results support the validity of the Solow model or the
augmented Solow model depending t on the samples that we investigated. The results also
revealed that FDI positively affects the growth rate of income per capita, however, not
significant for most of the samples. After all, there is no doubt that FDI is beneficial to the host
economy and governments should work on their policies for their countries to be the destination
of multinational corporations’ investment.
vii
Texas Tech University, Kolthoom Alkofahi, May 2014
LIST OF TABLES
Figure 2: FDI trends of world, developed and developing countries ....................................................... 34
Table I.A: OLS estimation of the Textbook Solow model – MRW samples............................................... 48
Table I.B: OLS estimation of the Textbook Solow model- new constructed samples ............................... 50
Table II.A: OLS estimation of the Augmented Solow model-samples of MRW ........................................ 57
Table II.B: OLS estimation of the Augmented Solow model-Developing countries and subsamples ........ 59
Table III.B: test for unconditional convergence, cross-sectional approach .............................................. 68
Table IV.A: Single cross-section results of conditional convergence, samples of MRW ........................... 72
Table IV.B: Single cross-section results of conditional convergence; Developing and sub developing
samples ................................................................................................................................................. 74
Table IV.B. Continued ............................................................................................................................ 75
Table V.A: Single cross-section results of conditional convergence Augmented Solow model. ............... 79
Developing and sub developing samples................................................................................................ 79
Table V.A. Continued ............................................................................................................................. 80
Table V.B: Single cross-section results of conditional convergence Augmented Solow model................. 83
Developing and sub developing samples................................................................................................ 83
Table V.B. Continued ............................................................................................................................. 84
Table VI.A: panel regression analysis, Non-Oil Sample ........................................................................... 93
Table VI.B: Results of panel regression analysis, Intermediate Sample. .................................................. 96
Table VI.C: Results of panel regression analysis, OECD Sample............................................................... 98
Table VI.C. Continued ............................................................................................................................ 99
Table VI.D: Results of panel regression analysis, Developing countries. ................................................ 101
Table VI.D. Continued .......................................................................................................................... 102
Table VI.E: Results of panel regression analysis, High income developing countries. ............................ 105
Table VI.E. Continued .......................................................................................................................... 106
Table VI.E: Results of panel regression analysis, Middle income developing countries. ........................ 108
Table VI.E. Continued .......................................................................................................................... 109
Table VI.F: Results of panel regression analysis, Low income developing countries. ............................. 111
Table VI.F. Continued .......................................................................................................................... 112
TABLE VII.1: Test for conditional convergence: Non-oil sample ............................................................ 126
Table VII.2: Test for conditional convergence: Intermediate Sample ................................................... 128
Table VII.2. Continued ......................................................................................................................... 129
Table VII.3: Test for conditional convergence: OECD Sample................................................................ 130
Table VII.3. Continued ......................................................................................................................... 131
Table VII.4: Test for conditional convergence: Developing Sample ....................................................... 138
Table VII.4. Continued ......................................................................................................................... 139
Table VII.5: Test for conditional convergence: High Income Developing Sample .................................. 140
Table VII.5. Continued ......................................................................................................................... 141
Table VII.6: Test for conditional convergence, Middle Income Developing Sample ............................... 142
viii
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.6. Continued ......................................................................................................................... 143
Table VII.7: Test for conditional convergence, Low Income Developing Sample ................................... 144
Table VII.7 ........................................................................................................................................... 145
Table VII.8: Restricted Conditional convergence using Pooled OLS ....................................................... 148
Table VII.9: Restricted Conditional convergence using fixed effects ..................................................... 148
Table VII.10: Restricted Conditional convergence (augmented model)................................................. 149
Table VII.11: Restricted Conditional convergence (augmented model)................................................. 150
Table VII.12: Restricted Conditional convergence (augmented model)................................................. 150
OECD sample ....................................................................................................................................... 150
parameter ........................................................................................................................................... 150
Pooled OLS .......................................................................................................................................... 150
Fixed effects ........................................................................................................................................ 150
Table IX. i : list of
coefficients based on the basic Solow model (restricted regression)............ 160
Table IX. ii : list of
coefficients based on Augmented Solow model (restricted regression) ...... 160
TableIX.iii: GMM estimations CEL and BHT, dependent variable is
............................................... 161
TABLE IX.1 GMM Estimation: Non-Oil Sample ...................................................................................... 167
TABLE IX.2 GMM Estimation: Intermediate Sample ............................................................................. 168
TABLE IX.3 GMM Estimation: OECD Sample ......................................................................................... 169
TABLE IX.4 GMM Estimation: Developing Income Sample .................................................................... 170
TABLE IX.5 GMM Estimation: High Income Developing Sample ............................................................ 171
TABLE IX.6 GMM Estimation: Middle Income Developing Sample ........................................................ 172
TABLE IX.7 GMM Estimation: Low Income Developing Sample ............................................................. 173
Table X.1: COUNTRIES IN THE STUDY SAMPLES OF MRW ..................................................................... 182
Table X.1. Continued ........................................................................................................................... 183
Table X.1. Continued ........................................................................................................................... 184
Table X.2: DEVELOPING COUNTRIES AND SUB-SAMPLES ...................................................................... 185
Table X.2. Continued ........................................................................................................................... 186
ix
Texas Tech University, Kolthoom Alkofahi, May 2014
LIST OF FIGURES
Figure 2: FDI trends of world, developed and developing countries .......................................... 34
x
Texas Tech University, Kolthoom Alkofahi, May 2014
CHAPTER I
INTRODUCTION
Foreign Direct Investment (FDI) started to play a significant role since 1980s
where both Developed and developing countries have started to attract significant
amount of FDIs. Economist, researchers and policy analysts have given considerable
attention to the relationship between economic growth and Foreign Direct Investment
(FDI). This relationship has been intensively studied in the literature and yet it attracts
no less attention today than it did any time before because of its anticipated spillovers
on productivity and economic growth. FDI is considered as a stable development
engine, especially in the developing countries, since these nations may lack the
knowledge and technology to utilize their resources efficiently and effectively. The
rationale behind it is that FDI may serve as a substitute to expose these economies to
new technologies and intellectual capital, which will then lead to economic growth.
The developing literture emphasizes technology transfers as a central aspect of
take-off and convergence of growth rate. Arguably, the most important channel of
technology transfer is FDI which is believed to boost economic growth irrespective of
whether an economy is developed or developing.
FDI that is carried by Multinational Corporations (MNCs) and Transnational
Corporations (TNCs) facilitates international technology diffusion, and affects
economic growth through three key mechanisms; size effects, skill and technology
effects and structural effects (Fortanier,F.; 2007).
The size effect is referred to as the net contribution of FDI to the host country’s
savings and investment, thus affecting the growth rate of the production base
(Bosworth and Collins, 1999).
The skill and technology effect is carried mostly by TNCs; it transfers skills
and technology across borders that decrease potential costs and increases benefits of
foreign capital (TNCs, 2007). Technology brought in by TNCs through FDI can spillover to local firms through demonstration effects, labor migration or linkages with
1
Texas Tech University, Kolthoom Alkofahi, May 2014
buyers and suppliers (Blomstrom, 1999). Local firms use the new technology to
increase their productivity and thus contribute to economic growth (TNCs, 2007).
The structural effects, on the other hand, brought about by the entry of a MNCs
and TNCs including both horizontal (competition) as well as vertical (linkage with
buyers and suppliers) changes. Investment of MNCs and TNCs can increase the
competition and improve the allocation of resources. The entry may contribute to the
innovation in the local market and thus to economic growth (Fortanier,F.; 2007).
Furthermore, according to the view point of neoclassical growth theory, FDI
inflows increase the stock of physical capital in host countries, thereby allowing
higher rates of growth than would be possible from reliance on domestic saving
(Youssef, Ali, 2001). FDI also increases the volume of investment and/or its
efficiency, and leads to long-term level effects and medium-term transitional increase
in growth (Usha Nair-Reichert and Diana Weinhold, 2001).
Moreover, Since there have been extensive inflows of FDI, it is important to
determine if that boom in FDI was beneficial to the economic growth, particularly of
developing countries. FDI may influence the rate of economic growth such that
conditional convergence occurs. Conditional convergence is the process that allows
countries to reach their steady state. Conditional income convergence is represented
by a higher percentage of Gross Domestic Product (GDP) growth for poor (and
middle) income countries towards their steady states compared to high income
countries. The technical diffusion from advanced economies to low income or
developing nations can partially explain conditional convergence (Donna Hak,2011).
Some researchers diverge from this view and argue that FDI is beneficial to
host countries depending on its characteristics, such as institutions (Alfaro et al.,
2004), liberalization (Anwara and Nguyen, 2010), openness to trade (Aviral Tiwari,
and Mihai Mutascu, 2011, and (Balasubramanyam et al., 1996), and technological
development (Borensztein et al., 1998). Nevertheless, FDI is being considered by
many as an important factor that helps in solving the problem of scarce local capital,
2
Texas Tech University, Kolthoom Alkofahi, May 2014
and overall low productivity in many developing countries (De Mello, 1999; Eller, et.
al, 2005). This means that the flow of foreign direct capital is argued to be a potential
growth-enhancing player in the receiving country (Al-Iriani, M.; Al-Shamsi, F.).
This article is motivated at first by the channels through which FDI contributes
to economic growth, and the role that FDI plays in the development of the economies,
as it may act as another factor of production (Shahbaz and Rehman 2010). For
example, FDI could affect economic growth through human capital accumulation,
which is considered among the key drivers of economic growth in developed and
developing countries (Sharma, B. and Gani, A.; 2004). Moreover, there is a lot of
articles emphasizing on the unidirectional or the bidirectional relationship that runs
from FDI to human capital or the other way around; In case where FDI involves
training of domestic labor, the strengthening of human capital will generate
externalities that could increase economic growth (Mustafa Akin and Valerica Vald;
2011).
This article is also motivated by the ongoing debate and ambiguity of the true
contribution of FDI to economic performance. While some literature support the
existence of a positive effect that runs from FDI to output and output growth
(Borensztein, 1998), De Mello (1999), and (Tiwari, A., Mutascu,M.; 2011), others
find negative or no such relation between FDI and economic growth; Kawai (1994),
Zukowaska-Gagelmann (2000), and Carkovic and Ross (2005).
The contreversial of the validity of the Solow model and whether to consider it
as adequate technique for growth literatures have also motivated me to do this work,
especially the widespread of endogenous growth literature that challenged, on
empirical grounds, the neoclassical growth theory of Solow model. Mankiw, Romer,
and Weil (MRW, 1992) for example, tested the validity of the Solow model in
explaining income differences across countries using Ordinary least square (OLS)
technique. MRW rejected the Solow model version in favor of the augmented Solow
model. On the other hand, (CEL,1996) used dynamic panel data approach in the form
3
Texas Tech University, Kolthoom Alkofahi, May 2014
of first difference generalized method of moments and rejected both the Solow model
and its augmentation with human capital. For these reasons, I reconsidered the
empirical case for the Solow model and used the work of MRW as a first step toward
my target to assess the adequacy of the Solow model and (or) its augmentation. (and
the true effect of fdi on the level of output.)
The reason behind our interest in the MRW in general, and FDI in particular,
falls in different pronged: first, MRW analyze the validity of the Solow growth model
and if its augmentation with human capital improves the fit of the model. Intuitively,
in the standard Solow type growth model, FDI enables host countries to achieve
investment that exceeds their own domestic saving and enhances capital formation.
According to this theory, the potential beneficial impact of FDI on economic growth is
confined to the short run. In the long run, given diminishing marginal returns to
physical capital, the recipient economy could converge to the steady state growth rate
as if FDI had never taken place leaving no permanent impact on economic growth of
the economy (De Mello, 1997). On the other hand, endogenous growth models that
highlight the importance of improvement in technology, efficiency, and productivity
suggest that FDI can positively influence the growth rate as it generates increasing
returns in productivity via externalities and production spillovers.
For the purpose of my research, taking advantage of this augmentation is what
I’m up to. The expansion of the Solow model using FDI allows me to check if results
are in parallel to what is generally thought about the true effect of FDI on the level of
income. It also helps me to attempt to answer the question that has attracted so much
attention in recent studies; whether per capita income and economic growth in
different countries or among different regions are converging or not. Many have found
evidence of convergence (Mankiw, Romer and Weil),(Alfaro;L, April 2003), and
(Villa Kaitila; 2004), others have been unable to do so (Charles,A., Darne,O., and
Hoarau, JF. 2009) and (Carkovic, M.; Levine, R., 2005).
4
Texas Tech University, Kolthoom Alkofahi, May 2014
MRW argued that countries are converging to their respective steady states at a
very slow rate, whereas it converge faster at a rate of 2% a year when the model is
augmented with human capital accumulation. This rate of convergence is akin to
Robert Barro (1991), and Xavier Sala-i-Martin (1991, 1992, 1995) who demonstrate
that this rate ranges from 2% to 3%. MRW findings contrasts to CEL(1996) where
they find evidence of faster convergence at a rate of approximately 10%, Somesh
Mathur (2005) as well finds evidence of conditional convergence ranging from 0.2%
to 22% in a year. The work of MRW helps us cast light on whether countries are
converging, their rate of convergence, and whether FDI helps countries to converge
faster to their respective steady states.
This study attempts to ask questions from two different angles; the methodology and
the consequences of incorporating FDI in the Solow model.
Can we still accept the Solow model as an adequate technique in explaining
income differences across countries; using new revised and extended data?
Consequently, How far the results produced by this work are different from those of
MRW? What would be the effect if we expand the Solow model to include FDI? In
other words, Can we consider FDI as a determinant of economic growth that may
narrow income per capita gap between rich and poor countries? What kind of effect
Does FDI exerts; positive, negative, or an insignificant effect? Does grouping
countries matter? Using conditional convergence, does the data support the Solow
growth model? Is there any evidence of convergence? At what rate? Can FDI
accelerate economic growth? Do we get different results by applying panel data
methodology, or using dynamic panel approach in the form of generalized method of
moments (GMM) estimation? Would it be of a great deal for countries to focus on
polices that attract more FDI?
This article uses data from two different sources; Penn World Table and World
Bank Table (WBT) to analyze the contribution of FDI to the empirics of economic
growth, during the period 1980-2010. The reason why this period is of a special
5
Texas Tech University, Kolthoom Alkofahi, May 2014
interest is that, as shown in figure I below, FDI grew very considerably over the past
three decades. This growth is due to the increase in the process of globalization, and
its activities that is mainly carried by the multinational and transnational corporations
(Fortanier,F.; 2007).
This article contributes to the existing literature in that, it takes MRW work as
its starting point ,and examines how the results change with the adoption of new
samples, data, and techniques. More specifically, this paper reevaluates the results of
MRW that aim to address the validity of the Solow growth model, using new, revised
and extended data, and new samples of countries. MRW study consists of three
samples; accordingly, this article added the Developing countries sample as an
extension to the existing samples. We further subdivided the developing countries
sample into three subsamples; high income developing country, middle income
developing country, and low income developing country.
Another contribution to this article ought to investigate a new extension to the
MRW study that , based on my knowledge, can’t be found in any previous literature;
while most literature focus on augmenting the Solow model with human capital
accumulation, this study append FDI as a potential factor that could explain
income differences across countries. This augmentation could refute some claims
about the illegitimacy of the Solow model, specially, since the endogenous growth
models embrace a diverse body of theoretical and empirical studies in the last three
decades.
The last contribution of this article is the application of a comprehensive study
of the textbook Solow model and its augmentation using three different techniques;
cross sectional estimation, panel data analysis and dynamic panel data technique in the
form of Generalized Methods of Moments. The advantage of conducting these
techniques is mainly threefold. First, draw an analogy to other literature inspired by
the work of MRW. The second advantage lies in the possibility of correcting any
biasness that might arise from using any of the techniques. Finally, textbook Solow
6
Texas Tech University, Kolthoom Alkofahi, May 2014
model and its augmentation together provides more robustness to the empirical results;
prevail as a solid defense against claims of invalidity of the Solow model, and backup
our prospects of FDI role in enhancing economics performance.
In this article, I will show that the net inflows of FDI support some of Solow
model’s aspects and fail in others. I will also show that for most of the samples, FDI
exert a positive effect on output per worker, and therefore on its growth rate. Although
the effect is not significant and not homogenous across countries in some samples
employed across sectional frame work, the results are positive and significant when
panel data analysis has been utilized. Furthermore, results show that the augmented
Solow model and its implications are satisfied to some degree. Finally, using the
conditional convergence, I will show that FDI accelerate economic growth mostly in
all samples
Before moving to the empirical analysis, this article reviews in details the
literature about the possible trace of FDI on level of income per worker and economic
growth, where the mechanisms and the empirical outcomes of the processes are
discussed. The data collections, samples, methodology, and estimation techniques are
explained in sections 3 and 4 respectively. Section 5 presents the results of the
analysis findings.
7
Texas Tech University, Kolthoom Alkofahi, May 2014
CHAPTER II
LITERATURE REVIEW
In this section I review some papers that describe the evolution of the literature
on the quantitative assessment of FDI effectiveness on output and economic growth
across countries and through time. Hence, this section is divided into three
subsections. First; Economic growth and the issue of convergence, where I briefly
present the main papers that introduced the methodology applied in the empirical
section of this paper. Second, FDI and human capital accumulation, I review in this
section the main papers that analyze the channels through which FDI and human
capital accumulations are related, the main reason to lay down this linkage is to justify
why we replace human capital accumulation with FDI. Finally, FDI and economic
growth, this part reviews the latest literature on FDI and its contribution to economic
growth. This extensive literature review hence provides the necessary information to
build the theoretical framework and methodological focus for this study.
ECONOMIC GROWTH AND THE ISSUE OF CONVERGENCE
Since the 1950s, economic-growth theories have evolved rapidly over time as
two distinct generations of models. The first generation of growth models is the
neoclassical (Solow or exogenous growth) model, as developed by Solow (1956),
Swan (1956), Cass (1965), and Koopmans (1965) with exogenous sources of longterm growth.
The second generation of growth models is endogenous-growth (the new
growth) model advanced with the theory of Romer (1986) and Lucas (1988). Romer
and Lucas highlighted the fact that technology change is an endogenous initiative
instead. These models focus on economic growth rate as a result of rational and
optimal agent’s behavior, and the structural characteristics of the economy. Since the
8
Texas Tech University, Kolthoom Alkofahi, May 2014
neoclassical side of the coin is of our interest, literature of endogenous growth model
is being cast a way.
In the neo-classical growth model, per capita growth rate is inversely related to
the initial income level, meaning that poor economies grow faster than advanced
economies, leading to conditional income convergence. The goal of the neo-classical
model is to predict conditional convergence, though not necessarily absolute
convergence. The empirical convergence literature starts with Abramovitz (1986) and
Baumol (1986). Abramovitz (1986) develops the hypothesis that the richest countries
converge while the world as a whole does not. Further research by Barro (1991) and
Mankiw et al. (1992) show the presence of conditional convergence which is the
ability of an economy to converge to its own steady-state. In general, many studies
have confirmed the presence of income convergence (Baddeley, 2006; Dawson and
Sen, 2007). Even though countries appear to approach their own steady states at a
fairly uniform rate of roughly 2 percent per year (Barro and Sala-i-Martin, 1991),
other studies find that countries may converge to its steady states at a higher rate of
convergence.
For more details on economic growth and the issue of convergence, I introduce
some influential literature and a short review of their impressive work.
Solow (1956) provided a methodology that is considered the first attempt to
model long-run growth analytically; it assumes that countries use their resources
efficiently , and assumes that there are diminishing returns to capital and labor. From
these two premises, the model makes three important predictions. First, increasing
capital relative to labor creates economic growth, since people can be more productive
given more capital. Second, poor countries with less capital per person will grow
faster because each investment in capital will produce a higher return than rich
countries with ample capital. Third, because of diminishing returns to capital,
economies will eventually reach a point at which any increase in capital will no longer
create economic growth. This point is called a "steady state".
9
Texas Tech University, Kolthoom Alkofahi, May 2014
Solow proposed in his classical article that, taking the growth rates of saving,
and population growth rate as exogenous, the steady–state level of income per capita
is determined by these exogenous variables, and since these variables vary across
countries, different countries reach different steady states. That is, the higher is the
saving growth rate (or the lower is the population growth rate) the richer is the
country. One of the Model’s implications is the elasticity of output with respect to
capital accumulations is equal to one third and two third with respect to labor. Another
implication is that countries that are not in their steady state are converging to their
respective steady state at rate of speed range between 2% and 4%. More about the
Solow model is explained in details in the section describing the methodology.
MRW (1992) test the validity of the Solow model and whether it is consistent
with the international variation in the standard of living. The study is conducted using
three samples encompass large set of countries; Non-oil, Intermediate income, and
OECD samples. They applied cross-sectional analysis that covers the period of 19601985. MRW argue that the predictions of Solow model are consistent with the
evidence; saving and population growth affect income in the directions that Solow
predicted. Moreover, more than half of the cross-country variation in income per
capita is found to be explained by these two variables alone. However, there are some
pitfalls of the model; first, even though the model correctly predicts the directions of
the effects of saving and population growth, it fails to predict the magnitude. Second,
the estimated value of the elasticity of output with respect to capital are found to be
much higher and less in conformity with its commonly accepted empirical values.
MRW propose that, to better understand the relation between saving,
population growth, and income, one must go beyond the textbook Solow model.
Accordingly, they have used a broad definition of the concept of capital; that is capital
consists of physical capital as well as human capital. In order to implement the model
with human capital, MRW have assumed that countries at the end of the period are in
10
Texas Tech University, Kolthoom Alkofahi, May 2014
their steady states. MRW augment the model with human capital as a factor that
explains income differences across countries. This augmentation could lower the
effect of saving rate and population growth rate on income level, it also produce a
lower rate of output elasticity with respect to capital.
MRW defined Income convergence as the tendency of poor country to grow
more rapidly than rich country. Using conditional convergence; poor countries grow
faster than rich countries allowing them to catch up with rich countries, they refute
what advocates of endogenous growth models claimed about the invalidity of the
Solow model. The evidences indicate that, holding population growth and capital
accumulation constant, countries converge at about the rate the augmented Solow
model predicts. More specifically, the speed of conditional convergence in their
samples ranges from 0.5% to 2% a year. He states that countries that are poor relative
to their own steady state do tend to grow more rapidly.
Islam (1995), on the other hand, follows the work of MRW, employs their
samples and set of data. Using cross-sectional analysis, both in unrestricted and
restricted forms, Islam analyzes the relationship between growth of output, initial level
of income, saving rate and the growth rate of population. He also augments the model
with human capital accumulation where the central focus of his study is the issue of
convergence. His work reaches to similar results as MRW; both, the effect of saving
and population growth rate on income are larger in absolute value, and the elasticity of
output with respect to capital are found to be much higher than proposed by the Solow
model. Moreover, the relationship between initial level of income and subsequent
growth rates for all samples are found to be negative which indicates the existence of
convergence even without incorporating human capital accumulations. However, his
results confirm the findings of a very slow rate of convergence.
Therefore, Islam suggests that there is sort of biasness produced by the data
or/and the methodology. Accordingly, Islam criticized one of the Solow model’s
assumption; identical aggregate production function for all the countries. Since it is
11
Texas Tech University, Kolthoom Alkofahi, May 2014
econometrically difficult and not easily measurable to allow for differences in the
production function using cross country frame work, he suggests that one should relax
this assumption to allow for such differences using another methodology to correct for
this bias.
As a first attempt, Islam advocates and implements a panel data approach to
deal with this issue; the omitting variable bias. Basically, he divides the full period
into shorter periods of five year spans, and takes the average of saving rate and
population growth rate over shorter periods for all samples and analyzes the
conditional convergence; that is convergence after differences in the steady states
across countries have been controlled for.
The usefulness of the panel data framework is that, it makes it possible to
allow for differences in the production function in the form of unobservable individual
country specific effect. However, the results show that dividing the period into shorter
spans and considering the growth process over shorter consecutive intervals does not
affect the results using cross pooled panel regression estimation; the study finds very
low estimates of the rate of convergence (0.6%-1.6% a year), and very high estimates
of the elasticity parameter.
On the other hand, results are robust when fixed effect panel estimations is
being employed; first, even without augmenting the model, there exist negative and
significant correlation between initial level of income and output growth. Second, the
elasticity of output with respect to capital is much plausible. Finally, the rates of
convergence are even higher than predicted (3.7%-9.1% a year).
Inclusion human capital has striking results; similar results to MRW are found
when applying single cross-sectional regression framework, the inclusion of human
capital does lower the elasticity of output with respect to capital, and lead to higher
rates of convergence. But when applying panel data estimates, coefficients of human
capital for all samples are negative and insignificant. Moreover, the inclusion of
12
Texas Tech University, Kolthoom Alkofahi, May 2014
human capital annihilates the effect that the cross section variation in human capital
had on the regression results.
Caselli, Esquivel, and Lefort(1996) Use large sample of developing and
developed countries over the period 1960-1985, they estimate a variety of growth
regression using different techniques; OLS, panel, and dynamic panel estimation in
the form of generalized method of moments. They criticize existing cross-country
empirical research on economic growth, showing that the statistical assumptions
underlying such work are violated. Moreover, they find that per capita incomes
converge to their steady state levels at a rate of approximately 10% a year. In another
application, a test of the Solow model and the augmented Solow model is tested.
Unfortunately, both models were rejected.
Bond, Hoeffler, and Temple (BHT, 2001), point out to some problems with
estimating growth regressions in general, and MRW methodology in particular. First,
the right-hand side variables are typically endogenous and measured with error.
Second problem that might arise is that of omitted variables. For example, the initial
level of efficiency, that is not observed, should be included in the right hand side of
the regression. Accordingly, this implies that the least squares parameter estimates are
biased.
BHT use two different Generalized Methods of Moments approaches; the firstdifferenced Generalized Method of Moments (1st-diff GMM), that was first introduced
by Caselli, Esquivel and Lefort (CEL, 1996). And a system Generalized Method of
Moments (SYS-GMM), suggested by Arellano and Bover (1995) and Blundell and
Bond (1998).
Using the MRW Non-oil sample, BHT apply OLS, within group, 1st diff and
SYS GMM estimations to the Solow and augmented Solow growth models.
Comparing their results to those of CEL, unlike the CEL convergence rate of 10% a
year, the system GMM results of Solow growth model indicate a rate of convergence
13
Texas Tech University, Kolthoom Alkofahi, May 2014
of around 2% a year, which is surprisingly similar to the standard cross-section
finding. However, the inclusion of human capital accumulation produces somewhat
more reasonable coefficient using 1-st diff GMM estimation.
After all, BHT demonstrates that, 1st-diff GMM estimator may be subject to a
large downward finite-sample bias, especially when the number of time periods
available is small. They also demonstrate that more plausible results can be achieved
using a (SYS-GMM) estimator. More about these methodologies is explained in
details in the methodology section.
It is worth mentioning that, the inclusion of human capital accumulation is not
the only determinant that researchers are incorporating in the Solow model. In fact, the
past decades have witnessed a renewed interest in the main factors driving economic
growth; researchers looked upon more determinants of economic growth that could
play significant role in explaining income differences across countries. For example,
the concept of capital in the neoclassical model can be usefully broadened from
physical goods to include human capital in the form of education, experience, and
health. Moreover (Barro R.,1996) found that the growth rate of GDP is enhances by
higher initial schooling and life expectancy, lower fertility, lower government
consumption, democracy, and lower inflation rate. In addition, others emphasize on
other macroeconomics factors that seem to play an important role for observed GDP
per capita patterns across countries. for example, (Bassanini and Scarpetta, 2001) shed
light on the important role of investment rate, Research and Development (R&D),
trade openness, well developed financial markets, Macroeconomic conditions and
policy settings in explaining income differences across countries.
In this paper, we aim at considering Foreign Direct Investment as possible
determinant of economic growth which we expect to play a significant role in
explaining income differences across countries, and possibly to enhance economic
performance and accelerate economic growth.
14
Texas Tech University, Kolthoom Alkofahi, May 2014
Somesh Mathur (2010) uses different forms of per capita growth equation to
test for conditional convergence hypothesis and also work out the speed of conditional
convergence for EU, East Asian and South Aisan regions together from 1961-2001.
Applying Solovian (1956) model, his finding emphasizes the existence of conditional
convergence among almost all pairs of regions except East Asian and South Asian
countries. The speed of conditional convergence ranges from 0.2% in a year to 22%.
He states that countries that are poor relative to their own steady state do tend to grow
more rapidly.
FDI AND HUMAN CAPITAL ACCUMULATION
Economists have long been emphasized on the relationship between FDI and
human capital accumulations. Some articles confirm the existence of a unidirectional
relationship that runs from FDI to human capital, while others find this relation is in
reverse. Never the less, few literatures reveal the existence of a bidirectional
relationship, and others uncover the negative relationship between these two economic
factors. The following articles shall highlight some aspect of these relationships.
Youssef, Ali (2001) tested the hypothesis that the level of human capital in
host countries may affect the geographical distribution of FDI; his empirical finding is
that, human capital accumulation is important and statistically significant determinant
of FDI inflows.
Sharma, B. and Gani, A. (2004) examined the effect of FDI on human
development for middle and Low-income countries. There results indicate that, FDI
exert a positive effect on human development for both groups of countries.
Mustafa Akin and Valerica Vald (2011) regress FDI on educational levels
across countries that are grouped based on Income. They find a positive relationship
between FDI and human capital in middle-income and upper middle-income countries.
On the other hand, an inverse relationship exists for the rest of the groups.
15
Texas Tech University, Kolthoom Alkofahi, May 2014
FDI, ECONOMIC GROWTH, AND CONDITIONAL CONVERGENCE
Foreign direct investment grew very considerably over the past three decades.
This notable growth triggered researchers to extensively study the contributions that
Multinational enterprises (MNEs) and FDI make toward economic growth of host
economies, especially developing countries. Developing countries lack the technology
and capital for investments and innovations. FDI allows these countries to have access
to advanced high tech products as well as technological, managerial, and intellectual
capital which helps in bringing them closer to their steady state. FDI may be
considered as an additional channel through which domestic economies can grow
faster (Zhang, 1999). Moreover, net foreign resource inflows can augment private
savings, and help countries reach higher rates of capital accumulation and economic
growth (Bosworth et al., 1999).
FDI per capita, FDI as a percentage of GDP, FDI inflows, and FDI net inflows
are often used as proxies for FDI. The different measures of FDI are used to check the
robustness of the regression analysis. Previous studies use one or more of the above
measures in order to determine the impact of FDI on economic growth. Ram and
Zhang (2002) introduce three variations of the FDI measure in their model, and find
that they yield similar results.
Attracting foreign capital inflows has become one of the prime policy goals in
transition economies, due to its growth-enhancing effects on the receiving economy.
The real GDP growth rates that the transition economies have experienced in the past
decade are above the world average (Sohinger, 2005). This confirms that these
economies are catching up to advanced economies.
However, little consensus have emerged as to whether FDI is boon or bane for
a country as a whole. Previous studies have come to conflicting conclusions regarding
the relationship between FDI, level of income per worker, and consequently economic
16
Texas Tech University, Kolthoom Alkofahi, May 2014
growth; the evidence is as mixed now as it has always been. Never the less;
tremendous claims about positive spillovers from FDI evidence are sobering.
The controversial regarding FDI mainly depends on the researchers’ analytical
view point; Microeconomics (industry-level), and Macroeconomics (the nation as a
whole). Since our study is a Macroeconomic based, only brief reviews of
Microeconomic pronged are included.
Microeconomic prospective
Microeconomic studies generally, though not uniformly, shed pessimistic
evidence on the growth effects of foreign capital. Some projects of particular countries
neither find evidence of FDI boosting economic growth, nor find positive spillovers
running between foreign and domestic sectors. On the other hand, other studies find
positive effects which create a debate:
Atkins and Harrison (1999), using panel data from Venezuelan plants,
uncovers considerable heterogeneity at the micro level. Although the study finds
foreign equity participation is positively correlated with plant productivity, this
relationship is robust only for small enterprises. It also finds no evidence of a positive
technology spillover from foreign firms to domestically owned ones between 1979
and 1989.
Haddad and Harrison (1993) explore the different consequences of FDI from
various countries of origin for economic growth in host countries. Using a group of
developed countries, they find the impact of FDI differs by country of origin.
Romer (1993), argued that important “idea gaps” between rich and poor
countries exists. He notes that foreign investment can ease the transfer of
technological and business know-how to poorer countries. According to this view,
FDI may boost the productivity of all firms including firms that are not receiving
17
Texas Tech University, Kolthoom Alkofahi, May 2014
foreign capital. Thus, transfers of technology through FDI may have substantial
spillover effects for the entire economy, and more specifically, on economic growth.
Alfaro, Chanda, Kalemli-Ozcan, and Sayek (2006), use an extended dataset,
find that the same amount of increase in FDI generates three times more additional
growth in financially well-developed countries than in financially poorly- developed
one.
Zukowaska-Gagelmann (2000) finds a negative impact of FDI on the
performance of the most productive local firms.
Reviewing these articles, we conclude that, based on the Microeconomic
prospective, FDI exert ambiguous effect on growth of output.
Macroeconomics prospective
Borensztein, De Gregorio and Lee (1995) develop endogenous growth
model, in which FDI increases long run growth through its effect on the rate of
technological diffusion from the industrialized world to the host country. They
conduct cross-country analysis of 69 developing countries, with panel data averaged
over two separate time periods 1970-79 and 1980-89. The dependent variables are percapita GDP growth rates over each decade. They conclude FDI, by itself, has a
positive but insignificant effect on economic growth. FDI is also an important
determinant of economic growth only when a country has a minimum threshold stock
of human capital. In that case, it actually contributes to growth more than domestic
investment does. Moreover, the authors find that FDI has the effect of increasing total
investment in the economy more than one for one.
Pradeep Agrawal (2000) presents empirical evidence on the impact of FDI
inflows on investment by national investors and on GDP growth. He employs OLS
time-series cross-sectional analysis of panel data (pooled regression) from five south
18
Texas Tech University, Kolthoom Alkofahi, May 2014
asian countries. This work finds the impact of FDI inflows on GDP growth rate is
negative prior to 1980, mildly positive for early eighties and strongly positive over the
late eighties and early nighties. These results provide some support for more liberal
policies toward FDI.
Usha Nair-Reichert and Diana Weinhold (2001) use mixed fixed effect and
random effect panel data estimation method. This method allows for cross country
heterogeneity in the casual relationship between FDI and growth. One important
finding of this study is that, the relationship between investment, both foreign and
domestic, and economic growth in developing countries is highly Heterogeneous. The
results indicate that there is considerable heterogeneity across developing countries
regarding the impact of FDI and other conditioning variables on economic growth.
Their results suggest that there is some evidence that the efficiency of FDI in raising
future growth rates, although heterogeneous across
countries, is higher in more open economies.
Khawar, Mariam (2005) uses cross country study to analyze the influence of
FDI from 1970 to 1991 on the growth of GDP per capita from 1970 to 1992, and
confirms the evidence of a strong positive correlation between FDI and growth of
GDP per capita. Another robust finding is that an increase in FDI leads to a relatively
large increase in GDP growth, especially when compared to other variables, for
example domestic investment.
Carkovic and Ross (2005) study the relationship between FDI and economic
growth using two econometric methods; a simple Ordinary least square (OLS) over
the period 1960-1995, with one observation per country. And a dynamic panel
procedure in the form of Generalized Method of Moments (GMM) with data averaged
over five-year period. The samples include all countries with available data during the
period of study .The study finds that the exogenous component of FDI does not exert a
robust, positive influence on economic growth. In the OLS regressions, initial income
19
Texas Tech University, Kolthoom Alkofahi, May 2014
and average year of schooling enter significantly and with the sign and magnitudes
similar to previous cross-country regressions. However, FDI is significant only for
some regressions. On the other hand, when a panel data analysis is applied, FDI enters
the regressions significantly for three out of seven regressions only under some
conditions. In Sum, their findings confirm that FDI is never significant in the OLS
regressions and becomes insignificant in the panel estimation when controlling for
financial development or international openness.
Aviral Kumar Tiwari, and Mihai Mutascu (2011) conducts an empirical
analysis in the framework of a panel estimation in order to analyze FDI-Growth and
Export-Growth nexus. The study also examines the impact of nonlinearities associated
with the relationship between FDI and growth, and exports and growth. The analysis
employs data from 1986 to 2008 for 23 aisan countries. The study finds that, both,
foreign direct investment and export enhance the growth process. However, exportled growth policies are more effective for growth enhancement of developing Asian
countries than FDI-led growth. In addition, labor and capital also play an important
role in the growth of Asian countries.
Syed Jawaid and Syed Raza(2012) investigate the relationship between FDI
and economic growth by using seven years average annual data of 129 countries from
the period of 2003 to 2009. Countries are further divided into three groups; Low
income, middle income, high income countries. Using OLS estimation technique, their
results indicate that, there exist significant positive relationships between FDI and
economic growth in all countries, as well as in all subsample countries. FDI contribute
more in Low income countries compared to other samples. Results of unconditional
convergence indicate that convergence exist in all country sample and all subsamples.
Results of conditional convergence confirm that middle and low income countries are
converging more rapidly in the presence of FDI.
It is worth mentioning that, Macroeconomic studies generally conclude that
FDI contributes to economic growth under certain circumstances. However, the
20
Texas Tech University, Kolthoom Alkofahi, May 2014
Macroeconomic findings in growth must be viewed skeptically, in the sense that
existing studies do not fully control for some potential problems; such as
methodological problems and a possible heterogeneity that might be hidden by the
data.
21
Texas Tech University, Kolthoom Alkofahi, May 2014
CHAPTER III
DATA AND SAMPLES
DATA
This section describes the data set that is utilized in this paper. In my first attempt,
and to be more in conformity to previous studies, I tried to collect data for the
variables of interest that cover the full period from 1960 to 2010. The problem that I
encountered in collecting data for all countries was the unavailability of all the data at
the beginning of the period for some countries. When constructing the samples similar
to MRW, missing data means that the samples are no longer similar, and hence,
comparing the results is somewhat problematic. For example, data for real Gross
Domestic Product per capita and FDI are not available before 1980 for a large number
of countries. On the other hand, FDI started to increase by the beginning of 1980s,
according to (Moosa, 2002), the 1980s witnessed some major changes that boosted
FDI inflows. This surge in FDI is attributed to the globalization of business and to
other factors that are explained later in details in section 5. If someone looks at figure
one below, he/she will notice that, the trends of FDI for the developing countries,
developed countries and the rest of the world are almost fixed and negligible before
1980. For these reasons, I look forward to shorten the period of interest to include data
that cover 1980-2010 only.
The basic data set used in this paper combines variables from two different
sources:
1. Version 7.1 of the Penn World Tables (PWT) that is described in Alan Heston,
Robert Summers and Bettina Aten, Nov 2012. PWT provides purchasing power
parity and national income accounts converted to international prices for 189
countries/territories for some or all of the years 1950-2010. It also displays a set of
national accounts economic time series covering many countries. The advantage of
22
Texas Tech University, Kolthoom Alkofahi, May 2014
using these tables lies in the structures of the data that allow us to compare real
quantity both between countries and over time.
I use PWT to extract Real Gross Domestic Product per Capita (RGDP per capita),
RGDP per worker, Population (in thousands), and Investment share of RGDP per
capita.
This study uses the growth rate of working age population that can’t be
found in PWT. The working age population is defined as the total population in a
region within a set range of ages that is considered to be able and likely to work.
The working-age population measure is used to give an estimate of the total
number of potential workers within an economy. Each region may have a different
range of ages, for that I unify this range to represent potential workers of age 15 to
64.
One way to find the working age population is to use RGDP per capita and
RGDP per worker. I use a simple math equation to find the total working age
population. First, I multiply the total population by RGDP per capita, which solves
for the total RGDP. Second, I divide the total RGDP by RGDP per worker and the
results are simply the total number of workers. From there I can find the growth
rate of working age population.
2. The World Bank Tables that include collection of development indicators,
compiled from officially-recognized international sources. It presents the most
current and accurate global development data available, and includes national,
regional and global estimates.
I use the World Bank Tables to borrow data for FDI that can’t be found in PWT.
FDI per capita, FDI % GDP, FDI inflows, and FDI net inflows are often used as
proxies for FDI. The reason behind choosing FDI % GDP in particular is because
all different measures yield to similar results (Ram and Zhang, 2002). All the data
I extract are annual and covers the period of 1980-2010.
23
Texas Tech University, Kolthoom Alkofahi, May 2014
Having all the data available for all the variables of interest, the dependent
variable varies depending on the equation we regress. For example, the log of GDP
per worker and its first difference are used in this paper’s regressions. The
independent variables, however, are as follows:
When the Solow growth model is applied, the log of investment share of
RGDP per capita (
) and the log of working age population
are used
as independent variables. The constant term (0.05) represent the depreciation rate and
the exogenous technology growth rate that is assumed to equal 2%. More about this
term is explained in the Methodology section. However, the log of net inflows of FDI
as a percentage of GDP
is included as independent variable, when the augmented
Solow growth model is applied instead. Moreover, when it comes to study the
conditional convergence and dynamic panel estimations, and to be more in conformity
with previous studies, I use the growth rate of GDP per worker
as the
dependent variable instead, and the log of initial level of GDP per worker
is
added as another explanatory variable.
SAMPLES
The samples I choose for the study are partly analogues to those of MRW. For
this reason, I first introduce a brief description for each sample and a clarification of
why certain countries are being selected or abandoned. I then present the new samples
that I construct with illustrations of why I choose these new samples.
In examining whether the Solow model is consistent with the international
variation in the standard of living, MRW consider three different samples. MRW first
sample is the Non-oil sample. It is the most comprehensive sample that consists of all
countries for which data are available, except those for which oil production is the
dominant industry. The reason why these countries are being excluded, according to
MRW, is because the bulk of recorded GDP for these countries represents the
24
Texas Tech University, Kolthoom Alkofahi, May 2014
extraction of existing resources, not value added. Accordingly, this sample in MRW
consists of 98 countries.
MRW second sample is the Intermediate income sample. It excludes countries
from the Non-oil sample whose data receive a grade “D” from Summers and Heston.
It also excludes countries whose populations in the 1960 were less than one million.
Summers and Heston use grade “D” to identify countries whose real income figures
are based on extremely little primary data. This sample of MRW contains 75
countries.
Finally, MRW choose the OECD countries to represent their last sample.
Countries that are member of OECD with populations of less than one million are
being excluded, for that the OECD sample consists of 22 countries.
I shall highlight that, the samples that I construct are slightly different than
those of MRW. Some countries consolidated as one country; for example, Germany.
Other countries disjointed to smaller countries; Russia. Zaire and Ivory Coast are
included with different names. Also, few countries emerged as dependent countries, or
existing countries emerged because of data availability. On the other hand, new
countries become a member of OECD by the year of 1980. Finally, the samples
exclude any country if it fails to provide FDI data at the beginning of 1980.
For these reasons, the samples that I construct analogous to MRW; the Nonoil, the Intermediate, and the OECD samples consist of 84, 74, 24 countries
respectively.
One of the extensions that this paper presents is the inclusion of new samples
in the study. These samples shall incorporate more structurally homogenous countries,
to avoid any measurement error that might be displayed as a result of heterogeneous
data. The best choice of samples to pick out is the developing countries sample. The
developing country is a nation with a low living standard, underdeveloped industrial
base, and low human development index relative to other countries.
25
Texas Tech University, Kolthoom Alkofahi, May 2014
To further narrow the differences between countries in this sample, I subdivide
this sample based on income as is classified in the World Bank Tables’ classification;
high income developed countries, middle income developed countries and low income
developing countries.
The reasons behind choosing these samples are because, traditionally, FDI was
a phenomenon that primarily concerned highly developed economies. In recent years,
however, the increase in FDI flows to developing countries turned out to be higher
than the increase in FDI flows to developed countries. As a result, developing
countries attracted almost half of world-wide FDI flows recently. This can be seen in
figure 1where it shows FDI trends for developing countries and the rest of the world.
Moreover, as explained in the literature review, some study emphasized that FDI plays
more important role in developing countries than in developed countries.
26
Texas Tech University, Kolthoom Alkofahi, May 2014
CHAPTER IV
AN OVERVIEW OF FOREIGN DIRECT INVESTMENT AND ITS
GLOBAL TREND
DIFINETIONS OF FOREIGN DIRECT INVESTMET
In a broad sense, FDI is composed of a flow of capital, expertise, and technology
into the host country. It is formally defined as "an investment made to acquire lasting
interest in enterprises operating outside of the economy of the investor”. Although it
has many definitions, the most widely accepted definition of FDI is known as “the
IMF/OECD benchmark definition” because it was provided by a joint workforce of
these two international organizations with the objective of providing standards to
national statistical offices for compiling FDI statistics:

According to the Detailed Benchmark Definition of Foreign Direct Investment,
Fourth Edition (BD4) of the OECD :Foreign direct investment reflects the
objective of establishing a lasting interest by a resident enterprise in one economy
(direct investor) in an enterprise (direct investment enterprise) that is resident in an
economy other than that of the direct investor.
The lasting interest implies the existence of a long-term relationship between the
direct investor and the direct investment enterprise. It also implies a significant degree
of influence on the management of the enterprise. The direct or indirect ownership of
10% or more of the voting power of an enterprise resident in one economy by an
investor resident in another economy is evidence of such a relationship. Some
compilers may argue that in some cases ownership of as little as 10% of the voting
power may not lead to the exercise of any significant influence, On the other hand, an
investor may own less than 10% but have an effective voice in the management.
27
Texas Tech University, Kolthoom Alkofahi, May 2014
Nevertheless, the recommended methodology does not allow any qualification of
the 10% threshold and recommends its strict application to ensure statistical
consistency across countries. Direct investment includes the initial equity transaction
that meets the 10% threshold and all subsequent financial transactions and positions
between the direct investor and the direct investment enterprise, as well as qualifying
FDI transactions and positions between incorporated and unincorporated fellow
enterprises included under the FDIR. Direct investment is not solely limited to equity
investment but also relates to reinvested earnings and inter-company debt (OECD
Benchmark, fourth edition, 2008).

According to the fifth edition of the IMF’s Balance of Payments Manual
(BPM5) defines FDI as a category of international investment that reflects the
objective of a resident in one economy (the direct investor) obtaining a lasting
interest in an enterprise resident in another economy (the direct investment
enterprise).
The lasting interest in a direct investment enterprise typically involves the
establishment of manufacturing facilities, bank premises, warehouses, and other
permanent or long-term organizations abroad, but may also involve the operation of
mobile equipment, such as drilling rigs, construction activities, and expenditures on
exploration for natural resources. This may involve the creation of a new
establishment abroad (Greenfield investments), joint ventures, or the acquisition of an
existing enterprise abroad (Merger and Acquisition). The direct investment enterprises
can be incorporated or unincorporated, and include ownership of land and buildings by
nonresident enterprises, as well as by nonresident individuals.
The lasting interest implies the existence of a long-term relationship between
the direct investor and the direct investment enterprise, and a significant degree of
influence by the investor on the management of the enterprise. A direct investment
relationship is established when the direct investor has acquired 10 percent or more of
28
Texas Tech University, Kolthoom Alkofahi, May 2014
the ordinary shares or voting power of an enterprise abroad. Further, in cases of FDI,
the investor’s purpose is to gain an effective voice in the management of the
enterprise. The foreign entity or group of associated entities that makes the investment
is called the "direct investor". The unincorporated or incorporated enterprise (a branch
or subsidiary in which direct investment is made) is referred to as a "direct investment
enterprise". Some degree of equity ownership is almost always considered to be
associated with an effective voice in the management of an enterprise; the BPM5
suggests a threshold of 10 per cent of equity ownership to qualify an investor as a
foreign direct investor. Once a direct investment enterprise has been identified, it is
necessary to define which capital flows between the enterprise and entities in other
economies should be classified as FDI. Since the main feature of FDI is taken to be
the lasting interest of a direct investor in an enterprise, only capital that is provided by
the direct investor either directly or through other enterprises related to the investor
should be classified as FDI. The forms of investment by the direct investor which are
classified as FDI are equity capital, the reinvestment of earnings and the provision of
long-term and short-term intra-company loans (between parent and affiliate
enterprises).
TYPES OF FDI
FDI can be classified from the perspective of the investor (the source country)
and from the perspective of the host country.
From the view point of the investor, FDI has three different types; Horizontal
FDI, Vertical FDI and platform FDI. Horizontal FDI arises when multi-plant firms
duplicate their home country-based activities at the same value chain stage in a host
country through FDI. On the other hand, Vertical FDI takes place when firms locate
different stages of production in different countries, i.e., when firms perform valueadding activities stage by stage in a vertical fashion in a host country. Finally,
29
Texas Tech University, Kolthoom Alkofahi, May 2014
Platform FDI in which the affiliate’s output is (largely) sold in third markets rather
than in the parent or host markets.
It is worth mentioning that the bulk of FDI is horizontal rather than vertical.
For example, the developed countries are both the source and the host of most FDI
suggesting that market access is more important than reducing production costs as a
motive for FDI.
From the perspective of the host country, FDI can be classified into: (i) Import
substituting FDI, (ii) Export-increasing FDI, (iii) Government initiated FDI.
MOTIVATIONS FOR FDI
There are many firms-specific motivations for why FDI is under taken; the
following outlines some of the motivations (Moosa, 2002):
Market Seeking: this kind of Investment targets either penetrating new
markets or maintaining existing ones. Resource Seeking Investments, on the other
hand, seek to acquire factors of production that is more efficient than those obtainable
in the home economy of the firm. Finally, Efficiency Seeking Investments in which
firms hope to increase their efficiency by exploiting the benefits of economies of scale
and scope.
FDI and SPILLOVERS
Many studies aim at analyzing the cost and benefits of FDI. Magnus Blomstrˆm &
Ari Kokko
30
Texas Tech University, Kolthoom Alkofahi, May 2014

Either break down monopolies and stimulate competition and efficiency or create a
more monopolistic industry structure, depending on the strength and responses of
the local firms;

Contribute to efficiency by breaking supply bottlenecks.

Introduce new know-how by demonstrating new technologies and training workers
who later take employment in local firm.
FACTORS THAT HELP BROADEN FDI
FDI is increasingly spreading throughout the world, some factors that
reinforced the widespread of FDI and capital includes:

Changing the economic policies.

Relaxing some of the restrictions on foreign sectors.

Offering of tax incentives and subsidies.

Lowering the barriers to trade and investment.

Other factors such as economic stability, the degree of openness of the host
Economy, income level, as well as the quality of institutions and level of
development might be thought to have connection to FDI inflows as well.
GLOBAL FDI TRENDS
The Global Foreign direct investment started to grow after 1980. In fact, before
that time, international trade was considered as the most important international
economic activity and grew more rapidly than FDI. This situation had changed
radically in1980s, with commercial banks lending to developing economies drying up,
most countries eased restrictions on FDI and many aggressively offered tax incentives
31
Texas Tech University, Kolthoom Alkofahi, May 2014
and subsidies to attract foreign capital (Aitken and Harrison 1999) and (World bank
1997a, 1997b). In the middle of 1980s, the world FDI started to increase its
importance by transferring technologies and establishing marketing and procuring
networks for efficient production and sales. Along with these policy changes and the
spillovers effects, it is widely believed that the process of globalization- that was
carried by Multinational Corporations (MNCs), Multinational enterprises (MNEs) and
Transnational corporations (TNCs) - mainly helped FDI getting its importance in the
world, along with the important role played by the noncommercial banks that helped
the private capital flows to developing economies in the 1990s. With the integration
of international capital markets, global FDI flows grew strongly in the 1990s at rates
well above those of global economic growth or global trade. Recorded global inflows
grew by an average of 13 percent a year during 1990-1997. Reaching a record US
$1.5 trillion in 2000, these inflows increased by an average of nearly 50 percent a year
during 1998–2000, and the increase was driven by large cross-border mergers and
acquisitions (M&A).
Data from the UNCTAD has shown that, the beginning of the FDI downturn
stared at the year 2001 mostly as a result of the sharp drop in (M&A) among the
industrial countries, coinciding with the correction in world equity markets. FDI
inflows continued to fall in 2002 reaching to its trough in 2003 ($729 billion).
Economists documented that the main factor behind the decline was slow economic
growth in most part of the world. Some other important factors were; falling stock
market valuations, lower corporate restructuring in some industries and the winding
down of privatization in some countries (UNCTAD 2003 World investment report).
Based on the data, the decline in FDI was uneven across regions, countries, and across
sectors. We can see this very obviously by looking at figure one that shows the world,
developed, and developing countries’ FDI inflows trends. We can tell that developing
countries were the least to be influenced by this fall in FDI.
32
Texas Tech University, Kolthoom Alkofahi, May 2014
Prospects after 2003 were promising; the world FDI started to recover when
the growth of global FDI continued to increase for four consecutive years. The global
FDI inflows rose in 2007 by 30% to reach $1.833 trillion, well above the previous alltime high set in 2000. All three major economic groupings- developed, developing
countries and the transition economies of some regions- saw continued growth in their
inflows. The increase in FDI largely reflected relatively high economic growth and
strong economic performance in many parts of the world. This increase is also
attributed to adopting more liberalization policies toward FDI inflows, combined with
other factors like; improved corporate profitability, higher stock valuations that
reflected higher profits return of foreign affiliates, notably in developing countries.
These recoveries of FDI lead to further recovery in International production that
carried out by transnational corporations (TNCs) in the developed countries as well as
the increase of TNCs in the developing countries.
Through the period of 2008-2009, the world economy suffered the deepest
global financial crisis since World War II. However, global foreign direct investment
flows rose moderately to $1.24 trillion in 2010; following the large declines of 2008
and 2009 but were still 15 percent below their pre-crisis average. In 2011, according to
UNCTAD, the Global foreign direct investment flows exceeded the pre-crisis average
reaching $1.5 trillion despite turmoil in the global economy. However, they still some
23 percent below their 2007 peak.
Data has shown that the largest flows of FDI occurs between the industrialized
countries (North America, Western Europe and Japan), it has also shown that flows to
non-industrialized countries are increasing rapidly. Recent data also revealed that the
United States is the world’s largest recipient of FDI. More than $325.3 billion in FDI
flowed into the United States in 2008, which is a 37% increase from 2007. On the
other hand, developing economies increased further in importance in 2010, both as
recipients of FDI and as outward investors. As international production and, recently,
international consumption shift to developing and transition economies, TNCs are
33
Texas Tech University, Kolthoom Alkofahi, May 2014
increasingly investing in both efficiency- seeking and market-seeking projects in those
countries. For the first time, they absorbed more than half of global FDI inflows in
2010. However, the observed uptrend in FDI was not evenly distributed among
different countries of the developing world. Half of the top-20 host economies for FDI
in 2010 were developing or transition economies.
Global trend of FDI
FDI inflows ( in billion $) of Global, Developed
and Developing economies, 1980-2010
2500
world
2000
Developed
1500
Developin
g
1000
500
0
1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010
Figure 1: FDI trends of world, developed and developing countries
34
Texas Tech University, Kolthoom Alkofahi, May 2014
CHAPTER V
METHODOLOGY, EMPIRICAL MODELS, AND DISCUSSION
OF RESULT
In attempt to come up with a successful study, the hardest part of all is to
choose a good model. Knowing that macroeconomics is not a one-size-fits-all type of
fields, it would be a daunting and complicated task to even attempt to construct a
model that explains all interesting macroeconomic phenomena; any such model will
make it difficult to learn, teach, and apply. For this reason, I will choose a model that
could help in explaining and answering the questions of interest, and is somehow
simple for the reader to comprehend.
The theoretical discussion of this article is built on that of the Solow growth
model (1956) and convergence hypothesis. Despite the fact that Solow’s purpose in
developing the model was to deliberately ignore some important aspects of
macroeconomics, it remains highly influential even today. Despite its relative
simplicity, it conveys a number of very useful insights about the dynamics of the
growth process.
The Solow model is worth teaching from a methodological perspective
because it provides a simple example of the type of dynamic model that is commonly
used in today’s more advanced macroeconomic theory.
To proceed, I would like to broadly discuss the basic textbook Solow model as
well as the augmented Solow model using the three econometric techniques; cross
sectional, panel estimation, and the generalized method of moments (GMM)
techniques, that will be discussed later in details.
The following shall highlight the main theoretical underpinnings of the Solow
model:
35
Texas Tech University, Kolthoom Alkofahi, May 2014
A. TEXTBOOK SOLOW GROWTH MODEL
In the Solow model, capital deepening is at the heart of the growth process.
There is only one commodity, output as a whole, whose rate of production is
designated
, thus
can be thought as the community’s real income. Part of
each instants output is consumed and the rest is saved and invested. It is worth
mentioning that the Solow model is built on the closed-economy version, where
Savings equals investments. This means that the additional to capital stock each period
depends positively on savings and negatively to consumption. The fraction of output
saved is constant, , so that the rate of saving is
.
The aim of the model is to explain the link between savings (s) and growth,
where savings are exogenous. At anytime, the economy has some amount of capital,
labor, and knowledge, and these are combined to produce output. These factors of
production are paid their marginal products that decline as more capital is
accumulated. The model also provides a useful framework for understanding how
technological progress and capital deepening interact to determine the growth rate of
output per worker.
The model assumes that GDP is produced according to an aggregate
production function technology that has the following general specification (The labor
augmented production function- or Harrod-neutral):
Where
is capital input,
is advancement in technology input, and
is
labor input.
It is worth flagging that most of the key results for Solow's model can be obtained
using any of the standard production functions that is seen in microeconomic
production theory. However, for concreteness, I am going to be specific and limit the
36
Texas Tech University, Kolthoom Alkofahi, May 2014
choice to the case where the production function takes the Cobb-Douglas form with
constant return to scale.
The initial level of capital
given.
and
labor
, and knowledge
are assumed to grow exogenously at rates
are taken as
and
, where
is defined as the growth rate of working age population for a specific country at time
, and
is defined as the growth rate of technology for the same country at time
.
What makes the Cobb-Douglas production function important in this study is its
two well-known features (assumptions) that are worth recapping:

Constant return to scale: (a doubling inputs leads to a doubling of outputs). In
other words, if a country X attempts to increase its total output, it could do so by
increasing the stock of capital and the number of workers are hired. Assuming the
country increased its inputs by a nonnegative constant c, then output changes by
the same factor (c). This can be explained mathematically by the following:
37
Texas Tech University, Kolthoom Alkofahi, May 2014

Decreasing marginal returns to capital accumulation: This turns out to be the key
element of the model. Adding extra capital (while holding labor input fixed) yields
smaller increase in output. For example, if a firm acquires an extra unit of capital,
it should raise its output. But if the firm keeps piling on extra capital without
raising the number of workers available to use this capital, the increase in output
will probably taper off. In particular, this can be seen by taking the second
derivative of output with respect to capital. For more demonstrations, I start with
the first derivative of output with respect to capital; the first derivative measures
the rate of change of output with respect to capital accumulations:
Equation (6) simply says that, capital accumulation positively affect the level
of output. This confirms that more capital input is necessary to increase the
production level, for sure to some degree.
The decreasing return to capital, on the other hand, can be seen by finding the
second derivative of output with respect to capital:
Equation (7) implies that the second derivative is negative since
.
Taking advantage of the Cobb-Douglas production function’s feature of
constant return to scale allows us to work with the production function in intensive
form:
Setting
in equation (5) yields:
38
Texas Tech University, Kolthoom Alkofahi, May 2014
The fraction
is defined as the amount of capital per effective worker. The
above equation is then reduced to:
Where:
is the level of output per effective worker,
effective worker, and
is the level of capital per
is the share of output that is devoted to capital accumulation.
The evolution of capital is the key equation for the Solow model, the
accumulation of capital per effective worker is governed by the following equation:
Where:
is the fraction of output saved,
is the rate of capital depreciation. Equation
(10) states that, the rate of change of the capital stock per unit of effective worker is
the difference between two terms, the actual investment per unit of effective
worker,
, and the break-even investment
; the level of
investment that is needed to keep the capital at its existing level. When the two types
of investment are equal, the country reaches to its steady state level of capital, where
the gain of extra capital is exhausted. Let’s refer to this level of capital as
Generally speaking, when a country starts with
state
.
below its level of steady
a positive net investment should be observed, which implies positive
growth of the stock of capital. If we consider a single country over time, the model
predicts that the growth rate will be high when capital per worker is low and will
decline as capital per worker rises (Inada condition). We have to bear in mind that, a
low value of capital per worker implies a high marginal product of capital which
means a high interest rate and a high level of investment. Therefore, we should
observe that the real interest rate declines along with capital marginal product as
economy develops. This movement to higher values of ( ) continues as long as
, where
is the steady state level of capital. In particular,
39
converges to
Texas Tech University, Kolthoom Alkofahi, May 2014
when the accumulation of capital is exhausted, in other word,
(k no longer
change over time).
To find the steady state level of output and capital per effective labor, we following
the argument that at the balanced growth path, the accumulation of capital is equal to
zero (i.e,
). Using equation (10) we find:
What equation (11) simplifies that, as the economy devotes more output to
investment, or if it experience lower population growth, higher steady-state level of
capital is achieved.
Similarly, the steady state level of output per effective worker can be found by
substituting equation (9) into equation (10):
The Solow growth model predicts that at the steady-state equilibrium, the level
of per capita income will be determined by the rates of saving, population growth, and
technological progress, all three of which are assumed to be exogenous. Since these
rates differ across countries, the Solow model yields testable predictions about how
differing saving rates and population growth rates might affect different countries'
steady-state levels of per capita income; other things being equal, countries that have
higher saving rates tend to have higher levels of per capita income, and countries with
higher population growth rates tend to have lower levels of per capita-income.
Moreover, assuming certain assumptions are satisfied, the process of within country
convergence towards the long-run equilibrium may result in a tendency towards
convergence in per capita income among economies.
40
Texas Tech University, Kolthoom Alkofahi, May 2014
In short, the Solow model implies that, regardless of its starting point, the
economy converges to a balanced growth path where each variable of the model is
growing at a constant rate.
The question that one might ask is at what rate this convergence occurs? More
specifically, how rapidly k converges to
? to measure the speed of convergence, or
how fast the stock of capital converges to its value at the steady state, we start by
reformulating equation (10) as a function of k, i.e.,
equals
. Notice that when k
, the accumulation of capital is equal to zero. Using a first order Taylor-series
approximation of
around k =
If we define
yields
, equation (13) reduces to
Equation (14) states that in the neighborhood of the balanced growth path,
moves toward
at a speed approximately proportional to its distance from
is approximately constant and equal to –
Moreover the growth rate of
this implies
Where
is the initial level of capital. To complete the simple textbook
Solow model, we need to find the (λ), we can derive (10) with respect to
evaluate the result at
. we get,
λ= -
41
and
)
.
,
Texas Tech University, Kolthoom Alkofahi, May 2014
Substituting equation (11) in the above equation yields:
Where:
is the share of income that goes to capital on the balanced growth path,
and is roughly predicted to equal one third.
Similarly, one can found that, output per unit of effective worker converges to
its steady state level at the same rate that the stock of capital per effective worker
converges (λ), that is
In sum, the Solow model has many important implications:
1. First of all, The Solow growth Model predicts that at the steady-state
equilibrium, the level of per capita income will be determined by the prevailing
technology (reflected by the production function), the rates of saving,
population growth, and technical progress, all are assumed to be exogenous.
Holding every other factor constant, countries with higher saving or/and lower
population growth rates tend to have higher levels of per capita income.
2. Savings rate do not affect the long-run growth of per capita income. The
crucial factor explaining the presence of a sustained long-run growth rate in an
economy is the presence of exogenous technological progress. However the
saving rate affects the long-run level of per capita income. It is only possible to
obtain continues growth in output per capita if there is exogenous technical
progress. In other word, the level of technology permanently affects the level
of output and stock of capital per effective labor at the steady state.
42
Texas Tech University, Kolthoom Alkofahi, May 2014
3. The Solow model predicted that the share of income that goes to capital (α) is
roughly one third, which means the elasticity of output with respect to capital
is one half.
4. Finally, as the Solow model predicts that α=1/3, the economies converge to its
steady state at a rate equal to 4% , where the key force that underlies the
convergence effect is diminishing returns to reproducible capital.
Despite the widespread use of the Solow model, there are some limitations that
worth mentioning:
First, the Solow model is built on the assumption of a closed economy. That is, the
convergence hypothesis supposes a group of countries having no type of interrelation;
in other word, no trade between countries has occurred. However, this difficulty can
be circumvented if we argue, as Solow did, that every model has some untrue
assumptions but may succeed if the final results are not sensitive to the simplifications
used. In addition to the model proposed by Solow, there have been some attempts at
constructing a growth model for an open economy, for example Barro, Mankiw, and
Sala-I-Martin (1995).
The second limitation is that the share of income with respect to capital does not
match the national accounting information. An attempt to eliminate this problem, as
done by Lucas (1988), involves augmenting the Solow model to include physical and
human capital; the latter consists of education and, sometimes, health.
The third limitation is that, the estimated convergence rate is too low even though
attempts to modify the Solow model have impacts on this rate; e.g., Diamond model
and open economy versions of the Ramsey-Cass-Koopmans model both have larger
rates of convergence.
Finally, the equilibrium rates of growth of the relevant variables depend on the rate of
technological progress taking into consideration that the individuals in the Solow
model (and in some of its successors) have no motivation to invent new good.
43
Texas Tech University, Kolthoom Alkofahi, May 2014
Accordingly, we attempt to reassess some of the model’s limitations (specially
the second and the third). We also test the validity of the Solow model using different
approaches to deal with some issues that the basic model ignores; such as correlated
individual effects and endogenous explanatory variables. On the other hand, some
literature emphasized that augmenting the Solow model with another factor, such as
human capital accumulation, show a better fit than the textbook Solow model.
Accordingly, investigating how robust the results are when the Solow model is
augmented with FDI will be the core of our study.
FIRST APPROACH: OLS CROSS COUNRY FRAMEWORK
In an effort to understand the quantitative relationship among saving,
population growth, and income, MRW modified equation (9) to see how differing
saving and labor force growth rates can explain the differences in the current per
capita income across countries. We first need to construct the steady state level of
output per worker (instead of output per effective worker) by multiplying both sides of
(9) by
Notice that the lower cases represent the per effective worker factor,
which means that multiplying these factor by
render these lower cases to reflect
per worker measures. For example:
Where:
is output per effective worker. However, if we multiply both sides
of equation (9) by
Where
, then:
is output per worker.
44
Texas Tech University, Kolthoom Alkofahi, May 2014
Taking logs of both sides of equation
the steady state income per worker
is:
The previous equation deserves attention. As MRW postulated, variation in
saving and population growth affect income in the direction that Solow predicted.
They have assumed that , the growth rate of technology, is the same for all countries,
and that
sum to 5% . On the other hand, MRW relied on a crucial assumption
when applied equation (18’) in their regression; the initial level of endowments of an
individual economy,
reflects technology resource endowments, climate,
institutions, etc, it may therefore differ across countries. Hence, they postulated that
, where
is constant and not a country specific, and
is the
country-specific shock term. According to them, since the rate of technology is
constant across countries, the term
can be dropped. Substituting
nto (18’)
yields
Equation (18) is our basic empirical specification in this section.
At this stage, MRW made the assumption that countries are at their steady
states at the end of the period. Moreover, MRW assume that
is independent of the
explanatory variables, s and n. This identifying assumption allowed them to proceed
with the Ordinary Least Squares (OLS) estimation. It is worth recapping here the
argument that MRW made to adopt the assumption of independency. First, this
assumption is made not only in the Solow model, but also in many standard models of
economic growth. Second, this identifying assumption renders it possible to test
various informal hypotheses that have been made regarding the relationship between
45
Texas Tech University, Kolthoom Alkofahi, May 2014
income, saving, and population growth. finally, since the specification above
postulates not only the signs of the coefficients but also their proximate magnitudes,
the regression results will allow testing of the joint hypothesis of validity of the Solow
model and the above-mentioned identifying assumption.
MRW used equation (18), where the log of output per worker at the end of the
period is taken to be the dependent variable
the log of investment share of RGDP per capita
population
. The independent variables are:
and log of working age
. Both independent variables are exogenous and averaged
over the full period 1960-1985.
DISCUSSION OF RESULTS
CASE I: SAMPLES OF MRW
Our first goal is to see how far the new revised and extended data are different
from those obtained by MRW. We employ equation (18) to see how different values
of and
can explain the differences in the current per worker income across
countries. We then test the implications of the restricted and the unrestricted Solow
model using cross-sectional approach. The restricted model reflects the assumption
that countries currently are at their steady states, which force the coefficients of
investment and the break-even investment to be similar in magnitude but opposite in
sign. MRW found the model to be quite successful in explaining a large fraction of the
cross country variations in income, but the estimates of the elasticity of output with
respect to capital were found to be very high. To know if the data and the samples we
construct generate similar outcomes, the dependent variable (
) is regressed on
log of investment share of RGDP per capita ( ), and log of working age
population
. The results of estimation equation (18) are reported in Table
I.A.
Table I.A includes the results of estimation for all the samples; the Non-oil,
Intermediate, and OECD samples. The first panel of the table gives results of
46
Texas Tech University, Kolthoom Alkofahi, May 2014
estimation in unrestricted form. The second panel, however, contains results of
estimating the equation after imposing the restriction; that is the coefficients of the
investment and population growth variables are equal in magnitude but opposite in
sign.
Like the results of MRW, there are some aspects of the results that support the Solow
model:
The coefficients on saving and working age population growth are highly
significant, and have the predicted sign for the Non-oil and Intermediate samples. For
the OECD samples, the coefficient of investment share of capital appears insignificant.
However, unlike MRW, the
Coefficients for the control variables are far from being equal; the estimated impact of
log of investment share on the log of income per worker are (1.458, 1.486, and 0.252)
for the Non-Oil, Intermediate and OECD samples respectively. On the other hand, the
estimated impact of the growth rate of working age population on the log of income
per worker, are in absolute values (4.682, 4.569, and 1.114). However, the
corresponding results of MRW show less difference between the corresponding
coefficients.
Moreover, the restrictions of coefficients are being equal in magnitude and
opposite in sign was not rejected by MRW for all samples, whereas our results reject
the null hypothesis for both the Non-oil and Intermediate samples.
Based on the unrestricted regression, we find that Large fraction of the cross
country variation in income per worker is due to differences in saving and working
age population growth. Our estimates of
for the Non-Oil, Intermediate, and OECD
samples are (0.50, 0.55, and 0.12), and the corresponding estimates of MRW
estimated as (0.59, 0.59, and 0.1). MRW found higher estimates for the Non-oil and
Intermediate samples, but unarguably lower estimates.
47
Texas Tech University, Kolthoom Alkofahi, May 2014
Table I.A: OLS estimation of the Textbook Solow model – MRW samples.
Note: regression
results of equation (18). Dependent
GDP per worker at 2010 ( OECD.) Numbers
Sample
Non-Oil variable is realIntermediate
in parentheses are t-statics.*, **, and *** denotes significance level at 1%, 5%, and 10% respectively
# of countries
(84)
(74)
(24)
Unrestricted regression
***
constant
***
-7.207
-6.850
***
7.099
(1.959)
(1.814)
(1.877)
1.458***
1.486***
0.252
(0.142)
(0.265)
(0.351)
***
***
-4.682
-4.569
-1.114**
(0.745)
(0.665)
(0.513)
0.50
0.55
0.12
0.96
0.83
0.31
-1.684
-1.719***
7.949***
(1.427)
(1.445)
(1.827)
1.982***
2.014
0.513
(0.254)
(0.255)
(0.308)
0.42
0.46
0.08
1.04
0.92
0.32
0.000
0.00
0.16
0.67
0.67
0.34
Restricted regression
Constant
Alpha
48
Texas Tech University, Kolthoom Alkofahi, May 2014
However, the finding of low values of
for the OECD may indicate that,
differences in income per capita across the sample is mostly attributed to variation of
technology. One might for the OECD sample. We can refer to the construction of the
samples and to the new revised data for this jump in the measurement of
might think that the low value of
.One
in both works is due to the insignificant
differences in income per capita across countries in that sample.
That last thing to discuss is the estimate of output share with respect to
. The estimates implied by the coefficients should equal capital’s share in
income
income that is implied by the Solow model, which is roughly (
However, our estimates of
).
equal to 0.67 for both the Non-oil and the Intermediate
samples, where as the corresponding estimates of MRW equal to 0.60. Both works
imply much higher value of
than implied by the national account information. It is
only for the OECD sample that the value of
is equal to 0.33, nevertheless, the
restricted model is being rejected which render this estimate to be of a less important.
After this analysis, we conclude that, the construction of samples similar to those of
MRW using new revised and extended data could neither produce better results, nor
support the aspects of the Solow model. Nevertheless, it is too early at this stage to
make a final judgment about the validity of this model in explaining income
differences across countries.
CASE II: DEVELOPING COUNTRIES AND SUBSAMPLES
The pitfalls of the results produced by Table I.A inspired us to construct a
sample in a way that the economies of participant countries share a lot of features. The
choice of
Developing income country is the best so far. According to the UN, a developing
country is a country with relatively low standard of living, undeveloped industrial
base, and moderate to low Human Development Index (HDI). We further classify this
49
Texas Tech University, Kolthoom Alkofahi, May 2014
Table I.B: OLS estimation of the Textbook Solow model- new constructed
samples
Sample
Developing all
High income
Middle income
Low income
# of countries
(64)
(14)
(20)
(30)
Unrestricted regression
constant
-0.830
2.265
7.861***
5.227
(2.968)
(2.225)
(2.138)
(3.988)
1.357***
1.488***
0.080
0.476*
(0.273)
(0.352)
(0.272)
(0.267)
-2.242
-1.287
-0.506
-0.554
(1.143)
(0.632)
(0.744)
(1.562)
0.31
0.56
0.08
0.05
0.94
0.33
0.38
0.67
Restricted regression
Constant
1.089
1.927
8.708***
5.409***
(1.431)
(1.869)
(1.451)
(1.341)
1.422***
1.455***
0.123
0.479*
(0.259)
(0.324)
(0.255)
(0.254)
0.32
0.60
0.08
0.08
0.94
0.32
0.38
0.65
0.46
0.81
0.60
0.96
0.59
0.59
0.11
0.32
Note: regression results of equation (18). Dependent variable is real GDP per worker at 2010 (
in parentheses are t-statics.*, **, and *** denotes significance level at 1%, 5%, and 10% respectively
50
.) Numbers
Texas Tech University, Kolthoom Alkofahi, May 2014
We run regression similar to that in Table I.A using the new constructed
sample. At this stage, we hope that the new choice of samples could produce better
estimates that are more inconformity with the Solow model’s implications. The results
of estimating equation 18 are represented in Table I.B.
Table I.B produce remarkable change in the results. Using cross-country OLS
approach, the results obtained for the developing countries sample are more plausible
than those of the Non-oil and intermediate samples. The coefficients of
and
are opposite in sign. The restricted model is not rejected at a very high
significance level. However, even though we found a lower estimate of α (0.59), this
value is considered very high relative to the value implied by the model.
One of the questions that we are willing to answer is whether or not the way
countries were grouped has any significant effects on the results of the regression.
According to Table I.B, sub classifying the full developing sample into three different
subsamples leads to a bitter fit of the model. Solow’s implications regarding
coefficients sign and magnitude are satisfied for the high income and low income
developing country. The coefficient of
is positive and equals to (1.488, and
0.476) respectively. Whereas the coefficient of
equals to (-1.287, and
-0.554) for the corresponding samples. On the other hand, the restricted regression is
not rejected for all the samples at more than 90% significance level. This means that
countries at the end of the period are at their respective steady state. Accordingly, the
estimated share of output with respect to capital is found to be very high for the high
income developing sample (α = 0.59), very low for the middle income developing
sample (α 0.11).it is only for the Low income developing sample that the value of α in
match to the value implied by the national account
information. Another aspect that supports the Solow model is that, differences is
saving and working age population growth for the developing and high income
51
Texas Tech University, Kolthoom Alkofahi, May 2014
developing country account for a large fraction of the cross-country variation in
income per worker.
According to the discussion of Table I.A and Table I.B, one can conclude that,
based on the cross-country regression, the Solow model’s implications are only met
for two of the samples; the OECD and the low income developing countries. This
could answer the question that we are considering; whether or not the way countries
are grouped affect the validity of the Solow model. Nevertheless, it is still too early at
this phase to say that the Solow model is inadequate and unsuccessful model in
explaining income differences across countries just because the estimated values of (α)
are considerably large. One way MRW suggest to reconcile the large value of α, is to
expand the textbook Solow model and include human capital accumulation as another
factor of input, and to see if this augmentation could lower the estimates of capital
shares of income, and could increase the fit of the model.
B. THE AUGMENTED SOLOW MODEL
Tables I.A and I.B produce results that are conditionally supportive of the
Solow model. The recognition of unusual high estimated values of (α) persuades
MRW to think of a necessary extension that should be made to the textbook Solow
model, which in turn leads not only to a better fit of the model, but also to more
realistic estimate of α for all the samples. One way to explain high estimates of α has
been to argue that capital and production function has to be understood in a broad
sense, so that the estimates obtained conform to the expected share of such broadly
defined capital in output. Another way to explain this finding is to consider the human
capital accumulation as a component of the error term in equation (18). Because
saving and working age population growth rates influence human capital
accumulation, one should expect human capital to be positively correlates with saving
rate and negatively correlated with working age population growth. Moreover, human
capital has been broadly quoted as principle engine for growth (Romer, 1986; Stokey,
52
Texas Tech University, Kolthoom Alkofahi, May 2014
1991, Khan, M., 2007), hence, excluding such a proxy can alter the analysis of crosscountry differences and create an omitted variable bias. This bias leads to over
estimation of the coefficients of independent variables, consequently, bias the
estimates of α.
It is so important to demonstrate how MRW augment their model with human
capital accumulation, since the central work of this study is to augment the Solow
growth model with FDI.
Taking advantage of the above argument, we are willing at using the extension
of the Solow model by employing FDI instead of human capital. The reason behind
choosing FDI in particular is the bulk of literature that emphasis on the significant
linkage between FDI and human capital accumulation (Blomstrom, M; Kokko,
A.;2007), (Eicher, T.; Kalaitsidakis, P.;1996), (Youssef, Ali,2001), and (Sharma, B.
and Gani, A.; 2004). On one hand, it has been theoretically proven that the effects of
human capital on growth and productivity, export promotion, technology transfers and
domestic economy have been significantly positive through FDI. On the other hand,
the evidence of various studies undertaken in countries that have developed human
capital reveals that human capital attracted FDI, subsequently, FDI impacted
positively on growth and productivity (Khan, M.; 2007).
Finally, based on the development literature that emphasizes on technology
transfers as a central aspect of take-off and convergence of growth rates, the most
important channel of technology transfer is found to be the foreign direct investment
(FDI). However, while theoretical models of FDI and firm location focus largely on
technology and physical capital, recent empirical evidence underscores that the
success of technology transfer via FDI depends crucially on the size of human capital
stock of the developing country (Borensztein, DeGregorio, and Lee [1995]), and
(Eicher, T.; Kalaitsidakis, P.;1996).
To keep the analysis manageable, I start with the Cobb-Douglas production
function:
53
Texas Tech University, Kolthoom Alkofahi, May 2014
Where, FI is the net inflows of FDI as a percentage of GDP averaged over the
full period of 1980-2010. All other variables have the same interpretations as in the
previous section.
At the methodological level, since FDI is a type of capital (the sum of long-term
capital; equity capital, and short-term capital); it depreciates at a rate of (
. One
change is being made to the assumptions; output shares with respect to physical
capital and FDI are assumed to be less than one (α
). This assumption implies
that, there is a constant return to scale in the reproducible factors, which ensures that
countries being at their respective steady state at the end of the period.
The dynamics of capital and FDI are represented by the following:
Where
and
are the fractions of income invested in physical capital and FDI.
The above equations imply that, the economy converges to a steady state
defined by
54
Texas Tech University, Kolthoom Alkofahi, May 2014
The above equations state that, the steady state level of capital and FDI depend
positively on the fractions of income that are invested in physical capital and FDI.
Substituting (21.1 and 21.2) into 19, and taking logs of both sides gives us an equation
for income per worker at the steady state:
This equation shows that income per worker totally depends on the control
variables; working age population growth, accumulation of physical capital, and the
net inflows of FDI.
At the empirical level, we run equation (22) using cross-country OLS framework, we
see how changes in the control variables could possibly explain income disparities
across countries of all samples of interest.
DISCUSSION OF THE RESULTS
CASE I: SAMPLES OF MRW
Table II.A reported the results of regressing equation (22), both in restricted
and unrestricted forms. The dependent variable is the log of income per worker at
2010. As in the textbook Solow model, the results fail to support all the Solow models
implications. Even though the coefficients of the log of saving rate and the log of
working age population predicted the right sign, the restricted model fails to reject the
null hypothesis for the Non-Oil and OECD samples. The last two lines in the table
give the values of α and
implied by the coefficients in the restricted difference
between the coefficients is getting even larger. On the other hand, incorporating FDI
55
Texas Tech University, Kolthoom Alkofahi, May 2014
as another factor of input positively affects the log of income per worker; however,
this effect is only significant for the Intermediate sample. FDI slightly improves the fit
of the regression for the intermediate sample; the estimated value of the
is now
0.62 compared to 0.55.
Based on the restricted regression, the assumption that countries at the end of
the period are at their respective steady states is not rejected for the Non-oil and
OECD samples. One of the objectives of this type of augmentation is to lower the
estimate of output share with the respect to capital. According to the data, the estimate
of α is lower but still higher than the generally acceptable value. Again, it is only for
the OECD sample that we can notice acceptable value of α. The parameter
is
estimated as (0.04, 0.12, and, 0.09) for the Non-oil, Intermediate, and OECD samples
respectively. For example, = 9% for the OECD sample, this can be interpreted as:
9% of income per worker of the OECD countries is devoted to foreign direct
investments activities.
The share of income with respect to FDI is a lot lower than the share with
respect to human capital, as represented in MRW Table II. This means that countries
are more interested in investing in human capital than in FDI, for it may contributes
more to economic growth than FDI does.
56
Texas Tech University, Kolthoom Alkofahi, May 2014
Table II.A: OLS estimation of the Augmented Solow model-samples of MRW
Sample
Non-Oil
Intermediate
OECD
# of countries
(84)
(74)
(24)
Unrestricted regression
***
***
**
-7.592
(1.980)
1.407***
(0.275)
-4.850***
(0.275)
0.161
(0.134)
0.51
-7.738
(1.696)
1.285***
(0.251)
-5.049***
(0.630)
***
0.440
(0.126)
0.62
5.672
(2.340)
0.469
(0.411)
-1.365**
(0.569)
0.095
(0.093)
0.12
0.95
0.77
0.31
-1.908
(1.452)
1.949***
(0.257)
0.131
(0.146)
-2.142
(1.395)
1.875***
(0.250)
0.384***
(0.139)
9.877**
(0.417)
0.738*
(0.365)
0.108
(0.095)
0.42
1.03
0.50
0.88
0.08
0.31
0.42
0.000
0.20
Alpha
0.63
0.58
0.40
Beta
0.04
0.12
0.09
constant
Restricted regression
Constant
Note: regression results of equation (22). Dependent variable is real GDP per worker at 2010 .
Numbers in parentheses are t-statics.*, **, and *** denotes significance level at 1%, 5%, and 10% respectively
57
Texas Tech University, Kolthoom Alkofahi, May 2014
In short, based on the cross-sectional analysis, augmenting the Solow model
with FDI is not of a great importance. FDI fails meet the Solow model implications
regarding the magnitude of the coefficients and the elasticity parameter.
Nevertheless, since FDI flows enormously to the Developing countries in
recent years, it is interesting to see whether FDI exert any positive and significant
effect on the Developing and sub-developing samples. Consequently, we use the
samples that we constructed to see whether the inclusion of FDI could support the
Solow’s implications, improves the fit of the model, and if it is considered as a major
factor that increase the level of income per worker across countries.
CASE II: DEVELOPING COUNTRIES AND SUBSAMPLES
Table II.B displays results of regressing equation 22 using the developing and
sub developing samples. The inclusion of FDI into the model could positively affect
the log of income per worker for the developing, high income developing, and middle
income developing samples, however, it is only significant for the high income
developing a=sample at 1% significance level. The coefficients of the control
variables are shown with the expected sing, and sum to zero for the high and low
income developing samples.
Moreover, even though the restricted model is not rejected for all the samples,
the estimated values of α remain considerably large for the developing and high
income developing samples. On the other hand, the estimated values of range
between 0.04 and 0.07.
is found to be the highest for the high income developing
countries, where it devotes 7% of its income per worker to be invested in FDI
activities.
58
Texas Tech University, Kolthoom Alkofahi, May 2014
Table II.B: OLS estimation of the Augmented Solow model-Developing countries
and subsamples
Sample
Developing all
High income
Middle income
Low income
# of countries
(64)
(14)
(20)
(30)
Unrestricted regression
constant
-0.983
(2.983)
1.625
(2.252)
7.678***
(2.245)
5.169
(4.087)
1.311**
(0.281)
1.314***
(0.353)
0.060
(0.284)
0.481***
(0.275)
-2.327***
(1.153)
-1.685**
(0.753)
-0.586
(0.791)
-0.575
(1.600)
0.112
(0.155)
0.181*
(0.059)
0.062
(0.163)
-0.019
(0.148)
0.30
0.62
-0.05
0.01
0.94
0.31
0.80
0.68
Restricted regression
0.979
(1.435)
1.378***
(0.265)
1.944
(2.231)
1.350***
(0.380)
8.617
(1.515)
0.110
(0.265)
**
5.431
(1.377)
0.486*
(0.263)
0.109
(0.154)
0.31
0.173
(0.099)
0.65
0.051
(0.158)
0.04
-0.018
(0.143)
0.05
0.94
0.29
0.38
0.67
0.46
0.82
0.57
0.95
Alpha
0.55
0.54
0.10
0.33
Beta
0.05
0.07
0.04
-0.01
Constant
Note: This Table reports the results from regressing equation (22). Dependent variable is
parentheses are t-statics*, **, and *** denotes significance level at 1%, 5%, and 10% respectively
59
***
. Numbers in
Texas Tech University, Kolthoom Alkofahi, May 2014
Unfortunately, Tale II.B uncovers a very important issue; the inclusion of FDI
negatively affects the low income developing countries. This unexpected results show
that inflows of FDI to this group harm the overall domestic income per worker. This
result can be attributed to the decrease in indigenous innovative capacity or crowding
out of domestic firms (Dunning,1988). Zengnaw A. Hailu, (2010) demonstrates that,
FDI may have negative effect if a country gives rise
to substantial reversal flows in the form of remittance of profits and dividends and/or
if the MNEs obtain substantial tax or other concessions from the host country.
Another way to interpret these results is that the expected positive spill-over
effects from the transfer of technology could be minimized because the technology
transferred is inappropriate for the host country’s factor proportion, especially; since
many developing countries have large agricultural sectors. Finally, an overly
restrictive intellectual property right might deter the inflows of FDI and produce
negative effect.
Over all, primarily conclusion can be drawn from the discussion above: Based
on cross-sectional analysis, utilizing recent and extended data produces results that are
inconformity with MRW’s findings. The results partially support the implication of
textbook Solow growth model. Surprisingly, adding FDI to the model does not
improve its performance for most of the samples. Therewith, it is too early to conclude
that both models are unsuitable in explaining income differences across countries just
because the results don’t perfectly support the model.
60
Texas Tech University, Kolthoom Alkofahi, May 2014
CHAPTER VI
THE ROLE OF FDI AND THE ISSUE OF CONVERGENCE
Despite the fact that recent growth theorists dismiss the Solow model in favor
of an endogenous growth model (that assumes constant or increasing returns to
capital), the different implications of exogenous and endogenous growth models have
lead to renewed empirical work in recent years. One of the major concerns has been
the issue of convergence. Conditional convergence is the tendency of poor country to
grow at a higher rate of income per worker growth than rich country and thereby
closing the gap between the two economies. Consequently, all economies should
ultimately converge in term of per capita income. Testing whether countries are
converging to their respective steady states is one way to support the Solow model.
Barro (1989) presented an argument that refutes the validity of the Solow
model. He quoted that: “convergence hypothesis seems to be inconsistent with the
cross-country evidence, which indicates that per capita growth rates are uncorrelated
with the starting level of per capita income”. In respond to this argument, Islam
(1995) declared that:” while finding evidence of convergence has been generally
thought of as evidence in support of the Solow-Cass-Koopmans model, absence of
convergence has been regarded as supportive of endogenous growth theories.”
Hence, our first goal in this regard is to reexamine this evidence on
convergence to assess wither it contradicts the Solow model. Second, the central focus
of our work is built on the way FDI affect the economy performance, therefore, we
examine the predictions of the augmented Solow model for behavior out of the steady
state. We hope to find a proof that FDI efficaciously strengthen the evidence of
convergence, subsequently, give more conformation of the legitimacy of the Solow
model.
The assumption MRW holds in their literature is that countries converged to
their respective steady state at the end of the period. This was followed by the crucial
61
Texas Tech University, Kolthoom Alkofahi, May 2014
assumption of the Solow model of diminishing marginal returns to capital. Intuitively,
the assumption of diminishing returns leads the growth process within an economy to
eventually reach the steady state where per worker output, capital stock, and
consumption grow at common constant rate equaling the exogenously given rate of
technological progress, and this lead to the notion of convergence. The concept of
convergence can be understood in term of level of income and in term of growth rate.
If countries are similar in terms of preferences and technology, then the steady state
income levels for all countries will be the same, and with time they will all tend to
reach that level of income per worker. On the other hand, since in the Solow model the
steady state growth rate is determined by the exogenous rate of the technological
progress, then provided that technology is a public good to be equally shared, all
country will eventually attain the same steady state growth rate.
The neo-classical theory distinguishes two types of convergence: unconditional
and conditional convergence. The unconditional convergence is when it is assumed
that all countries converge to the same steady state level of output. On the contrary,
convergence is said to be conditional, if countries are assumed to converge to their
respective steady states, or when differences in the steady states across countries have
been controlled for.
The traditional neo-classical approach in finding convergence in per worker
income is referred as “beta” convergence. It is obtained by a regression analysis
estimating the correlation between initial levels of income and subsequent growth
rates. Because of diminishing marginal returns to capital, countries with low levels of
capital stock will have higher marginal product of capital, and for similar saving rates,
it grow faster than those with higher levels of capital per worker. Thus finding a
negative correlation indicates that countries with a lower initial level of income per
worker grow more rapidly than countries with a higher initial level of per capita
income, and thus convergence holds in both terms; income level and growth rate.
62
Texas Tech University, Kolthoom Alkofahi, May 2014
There is conflicting evidence to whether FDI help accelerate or slowdown
convergence. For example, (Jawaid, Raza; 2012), based on conditional convergence,
found the results suggest that low and middle income countries are converging each
other more rapidly. (Changk Choi, 2004) investigated convergence in income level
and growth rate using panel approach for the OECD countries. He found that income
level and growth gaps between source and host countries turn out to decrease as
bilateral FDI increases. (Carkovic, Levine; 2005) found that FDI inflows do not exert
an independent influence on economic growth. Moreover, (Joze Mencinger, 2003)
finds a negative correlation between FDI and economic growth for the Baltic
countries; this means that FDI slowdown convergence toward the steady states.
My next goal is finding evidence of convergence for all the samples included
in the study using dynamic ordinary least square (DOLS). Finding evidence of
convergence is considered as a way to support the legitimacy of the Solow model. The
assumption of countries are being at their respective steady states at the end of the
period will be relaxed later in this section by considering out of steady states behavior.
Another goal in this regard is checking whether FDI is an important factor that helps
accelerate economic growth and hence, convergence. Finally, analyze whether FDI is
contributing more in low income developing countries, middle income developing
countries, or high income developing countries.
Empirically, the Solow model predicts that countries reach different steady states
(conditional convergence). It also makes quantitative predictions about the speed of
convergence (λ) to steady state. The way to find the speed of convergence is to
consider out of steady states behavior.
Following the analytical approach of MRW to study the rate of convergence,
we start by approximating around the steady. The speed of convergence is given by
63
Texas Tech University, Kolthoom Alkofahi, May 2014
Where
is the speed of convergence; is the speed at which actual income is reaching
its steady state level of income in a year, and has the following representation
Let’s define some variables of interest:
the steady state level of income per effective worker.
the actual level of income per effective worker at time t.
Log linearizes equation (23) yields:
Subtracting
from both sides, and substitutes for
, the above equation reduces
to
(
Where:
is the log difference of income per effective worker.
It is worth mentioning that
takes different values based on the estimation we follow.
For example, in case of cross sectional framework, =30. It refers to the difference
between the end and the beginning of the period; (1980-2010).
Thus, in the Solow model, the growth of income per worker is a function of the
determinants of the ultimate steady state and the initial level of income.
The left hand side measures the growth rate of income per worker over the
period 198-2018. If the coefficient of initial income is negative and significantly
different from zero, then data exhibits conditional beta convergence. This finding
implies that countries that are far from their respective steady states will grow at a
faster rate than countries that are closer to their respective steady states. In this section,
64
Texas Tech University, Kolthoom Alkofahi, May 2014
I would test for unconditional and conditional convergence, and work out the speed of
convergence using OLS estimation.
a) TESTS FOR UNCONDTIONAL CONVERGENCE
We now test the convergence predictions of the Solow model. Analytically,
two different test of convergence have been performed for all the groups. The results
of unconditional convergence for all samples are reported in Table III.A and Table
III.B, while the tests for conditional convergence are reported in Tables IV.
DISCUSSION OF RESULTS
CASE I: SAMPLES OF MRW
To test for unconditional convergence, the log difference of income per worker
is regressed on the initial level of income per worker. The test is being conducted for
all the samples and subsamples of the study. Table III.A reports the results of
unconditional convergence for samples analogues to those of MRW. The results carry
same implication as in MRW. The coefficient on the initial level of income per worker
is slightly positive but insignificant for the Non-Oil and intermediate samples, and the
adjusted
is almost zero, these results indicate that there is no tendency for poor
countries to converge faster than rich countries.
However, table III.A does show that there is a significant tendency toward
convergence in the OECD sample. The coefficients on the initial level of income per
worker is significantly negative, and smaller than that of MRW, the adjusted
of the
regression is 0.41 compared to
MRW estimate of 046, and the speed of convergence is 2% compared to 1.67% for the
same sample. However, both rates are found to be very low compared to the value
implied by the model of 4%. One way finding the rate of convergence is important, is
to predict how long it takes countries to reach half way toward their respective steady
65
Texas Tech University, Kolthoom Alkofahi, May 2014
states. For example, when λ=2%, it takes the OECD countries35 years to reach half
way toward their respective steady states. This, however, contradicts what implied by
the Solow model; where countries need 17.5 years to reach half way to the steady
states.
The reader may want to know how we found the estimate of λ especially, with
unknown value of α . Frankly speaking, this value is derived from the estimated values
of the initial level of income per worker.
For more illustration on how to find the value of λ, we start by the following equation:
Let’s denotes abbreviate to the initial level of income by . Where:
Rearrange this abbreviation, and take the log of both sides yields
This term is reduced to
Notice that τ is just a fixed number that represents the time period of the study. In case
of the cross sectional approach τ equals to 30 (2010-1980).
Hence:
Likewise, to determine how long it takes each year to reduce the income per worker
gap by half, one should measure the half life of convergence for each sample. It can be
calculated by the following:
66
Texas Tech University, Kolthoom Alkofahi, May 2014
Half life convergence
For example, if
0.05, then it takes
years for countries to reduce the
income gap by half.
Table III.A: test for conditional convergence, cross-sectional approach
Test for unconditional convergence
Dependent variable
: 1980-2010
Sample
Non-Oil
Intermediate
OECD
Observation
84
75
24
Constant
-0.189
(0.389)
0.078
(0.386)
5.082***
(1.131)
0.047
(0.041)
0.022
(0.040)
-0.440***
(0.107)
0.003
-0.009
0.41
s.e.e.
0.47
0.40
0.23
Implied λ (in % a year)
0.0
0.0
2.0
Half life of convergence
(in years)
35
Note: Numbers in parentheses are t-statics.
*, **, and *** denotes significance level at 1%, 5%, and 10% respectively
67
Texas Tech University, Kolthoom Alkofahi, May 2014
CASE II: DEVELOPING COUNTRIES AND SUBSAMPLES
Table III.B, on the other hand, does show considerably different results than
those of Table III.A. There exist a significant inverse relationship between the initial
level of income per worker and subsequent growth rates for the high income, middle
income, and low income developing samples. The log of initial level of income per
worker is found significant at 1% significance level. The highest value of the
coefficient is (in absolute value) is found for the high income developing sample (0.618), is found highly significant for the middle Income sample (-0.573), and the
lowest value is shown for the low income developing sample, where is significantly
estimated by (-0.237). These results show that, there are strong tendency toward
convergence for all sub-classified samples which in turn refute the claims of
illegitimacy of the Solow model.
Table III.B: test for unconditional convergence, cross-sectional approach
Test for unconditional convergence
Dependent variable
: 1980-2010
Sample
Developing
High income
Middle income
Low income
Observation
64
14
20
30
Constant
0.165
(0.582)
6.522*
(3.217)
5.502***
(1.175)
1.931
(1.451)
0.004
(0.066)
-0.016
-0.618*
(0.324)
0.17
-0.573***
(0.128)
0.50
-0.237**
(0.084)
0.02
s.e.e.
0.527
0.49
0.30
0.55
Implied λ (in % a year)
0.0
3.2
2.8
1.0
22
25
69
Half life of
convergence( in years)
Note: Numbers in parentheses are t-statics.
*, **, and *** denotes significance level at 1%, 5%, and 10% respectively
68
Texas Tech University, Kolthoom Alkofahi, May 2014
Moreover, the coefficient of adjusted
is considerably high for the middle
income developing sample, this assures that, the initial level of income per worker is
very informative in determining subsequent growth rates.
The last raw of the table includes number of years needed for income per worker to
double. For example, it takes income per worker of the high income developing
countries approximately 22 years to reach to half way toward its steady states.
An important issue that has arisen is that, the idea of convergence (sometimes
known as the catch-up effect) is the hypothesis that poorer economies’ incomes per
worker will tend to grow at faster rates than richer economies. As a result, all
economies should eventually converge in term of income per worker. Unfortunately
this is not interpreted by our data. The results show that the speeds of convergence for
high, middle, and low income developing countries equals to (3.2%, 2.8%, and 1%).
Clearly the highest value of λ refers to the high income developing sample and this
contradicts the hypothesis of conditional convergence. This means that the gap of
income per worker between the two economies is widen indefinitely, in other world,
this is manifest that the poor countries are get poorer, and the rich countries are getting
richer which is not conspicuously true.
Another issue in this regard is the failure of developing countries to converge.
It is widely believed that the developing countries have the potential to grow at a faster
rate than developed countries. This is true because, the diminishing returns (in
particular, to capital) are not strong as in capital-rich countries. This is also true
because poorer countries can replicate the production methods, technologies, and
institutions of developed countries. However, the results reported in the table do not
interpret this discussion. According to Jeffrey Sachs (1997), convergence is not
occurring everywhere because of the closed economy policy of some developing
counties, which could be solved through free trade and openness. On the other hand,
Moses Abramovits (2000) emphasized the need for 'Social Capabilities' to benefit
69
Texas Tech University, Kolthoom Alkofahi, May 2014
from catch-up growth. These include an ability to absorb new technology, attract
capital and participate in global markets. According to Abramovitz, these prerequisites
must be in place in an economy before catch-up growth can occur, and explain why
there is still divergence in the world today.
After all, we conclude that the data of our model reject the hypothesis of
unconditional (sigma) convergence. We now move to investigate the conditional
(beta) convergence predictions of the Solow model.
b) TESTS FOR CONDTIONAL CONVERGENCE
1. THE TEXTBOOK (BASIC) SOLOW MODEL
As previously mentioned, conditional convergence is defined as the existence
of an inverse relationship between initial level of income per worker and its
subsequent growth rates, once we control for the determinants of the steady state level
of income per worker. These determinants that we considered are the control variables
of the Solow model; the average annual savings rate, the rate of growth of working
age population, and the growth rate of technology that the model assumed is
exogenous. The inverse relationship reflects that countries that are poor relative to
their own steady state do tend to grow more rapidly.
DISCUSSION OF RESULTS
CASE I: SAMPLES OF MRW
Table IV.A represents the result of estimating equation (25) using restricted
and unrestricted regressions. MRW’s results of conditional convergence are available
only in the unrestricted form. However, Islam (1995) replicated the work of MRW and
included the Results in both restricted and unrestricted forms. For this reason, I started
the analysis by comparing the results of the unrestricted regression.
70
Texas Tech University, Kolthoom Alkofahi, May 2014
The upshot of Table IV.A pointed out to the fact that, the inclusion of
and
substantially improved the fit of the regression. The coefficient of
the initial level of income per worker is now significantly negative for all the samples
included; that is, there is strong evidence of convergence. On the other hand, a
comparison between results of Table IV.A and results of MRW and Islam shows that
our estimates of the coefficient of initial income per worker are lower for Non-oil and
intermediate samples but higher for the OECD sample. According to our data, this
means that, both Non-Oil and Intermediate samples are further away from their steady
states. This is also reflected in the respective speed of convergence for all three
samples; λ is equal to (0.005, 0.006, and 0.02) for the Non-oil, Intermediate, and
OECD samples. The corresponding estimate for λ in MRW and Islam is found
( 0.006, 0.010, and0.017) and (0.005, 0.010). it is clear that the Non-oil sample in this
study is converging at the same low rate as the other two studies. For the Intermediate
sample, λ is found to be lower than the corresponding estimates of MRW and Islam.
However, λ for the OECD is converging at a faster speed of convergence.
The inclusion of the log of the control variable of the Solow model to the righthand side of the regression, substantially improves the fit of the regression. The values
of
after the inclusion is (0.30, 0.94, and 0.62) compared to (0.003,-0.009, and
0.04). The corresponding estimates of MRW are (0.38, 0.35, and 0.62).
The results from the restricted estimation allowed us to get unique estimates of not
only λ, but also the output elasticity parameter, α .The estimates of λ obtained from the
restricted estimation are slightly different than those from the unrestricted model for
the Non-oil and the Intermediate samples. In general, they confirm the finding of a
very slow rate of convergence
On the other hand, the estimate of α is found to be 0.89 for the Non-oil sample,
0.85 for Intermediate, and 0.56 for OECD. These are unusually high values, and even
higher than Islam’s corresponding estimates for the Non-oil and Intermediate samples.
The corresponding estimates of α are unavailable for MRW, for that we only reported
71
Texas Tech University, Kolthoom Alkofahi, May 2014
Table IV.A: Single cross-section results of conditional convergence, samples of
MRW
Sample
Non-Oil
Intermediate
OECD
# of countries
(84)
(74)
(24)
Unrestricted regression
constant
***
-4.312
2.061
(0.810)
(0.700)
(1.542)
-0.113**
-0.157***
-0.466***
(0.044)
(0.041)
(0.099)
***
0.480
(0.121)
-1.586
(0.340)
Hal-life convergence (in years)
***
0. 553
(0.110)
***
Implied λ (in % a year)
***
-4.032
***
0.435
(0.367)
-1. 564
-0.758**
(0.291)
(0.342)
0.30
0.94
0.62
0.39
0.32
0.20
0.5
0.6
2.1
138
115
33
-2.385***
2.219
Restricted regression
Constant
-2.450***
(0.567)
(0.528)
***
***
-0.073
Alpha
Implied λ (in % a year)
Hal-life convergence (in
-0.111
(1.530)
-0.445***
(0.044)
(0.041)
(0.096)
0.601***
0.657***
0.492 **
(0.102)
(0.112)
(0.096)
0.91
0.93
0.623
0.41
0.34
0.20
0.00
0.00
0.35
0.89
0.85
0.56
0.3
0.4
2.0
231
173
34
years)
Note: Results of regressing equation 25. Dependent variable is
**, and *** denotes significance level at 1%, 5%, and 10% respectively
72
. Numbers in parentheses are t-statics.*,
Texas Tech University, Kolthoom Alkofahi, May 2014
the finding of Islam who also found a very high estimates of α as well. These
estimates are . Islam found high values of α of (0.83, 0.76, and 0.60).
CASE II: DEVELOPING COUNTRIES AND SUBSAMPLES
We now move to find out if results differ when the new constructed samples are used.
Table IV.B reported the results of the regression using the developing and all subdeveloping samples. These results are similar in the spirits to those of Table IV.A; the
coefficient of initial level of log income per worker is negative across all the
subsamples, however, it is only significant for high and middle income developing
samples.
This negative relationship signifies the inverse relationship between growth
initial level of income per worker and subsequent growth rates. Such results confirm
the presence of that conditional convergence among the subsamples, where countries
are converging to their own steady states level of per worker income.
The coefficients of average annual saving rate and average of growth rate of
working age population are significant for most of the samples. They support the
Solow implication
regard the sign but fails to be equal. Again, the speed of convergence is 2.9% and the
highest for the high income developing countries and is the lowest for the low income
developing samples and equals to 0.4%.
Without any doubt, including these control variables increases the fit of the regression
for all the samples. The implied coefficients of
for the unrestricted estimation are
(0.32, 0.72, 0.59, and 0.21) where the corresponding estimates of unconditional
convergence are (-0.016, 0.17, 0.50, and 0.02) respectively.
The restricted model fails to reject the null hypothesis at 5% significance level
for all the subsamples. Our finding confirms the failure of countries to converge to the
73
Texas Tech University, Kolthoom Alkofahi, May 2014
Table IV.B: Single cross-section results of conditional convergence; Developing and sub developing samples
Sample
# of countries
Developing all
(64)
High income
(14)
Middle income
(20)
Low income (30)
Unrestricted regression
constant
-6.140***
(1.409)
-2.392
(2.943)
0.708
(2.246)
-7.153*
(3.881)
-0.093
(0.059)
-0.583**
(0.189)
-0.469***
(0.124)
-0.119
(0.181)
0.469***
(0.138)
0.148***
(0.312)
0.148**
(0.238)
0.297
(0.200)
-2.245***
(0.525)
-1.496**
(0.668)
-1.318
(0.557)
-2.883**
(1.246)
0.32
0.72
0.59
0.21
0.43
0.28
0.27
0.49
Implied λ (in % a year)
0.3
2.9
2.1
0.4
Half life convergence
(in years)
230
24
33
173
74
Texas Tech University, Kolthoom Alkofahi, May 2014
Table IV.B. Continued
Sample
# of countries
Developing all
(64)
High income
(14)
Middle income
(20)
Low income
(30)
Restricted regression
-2.215***
(0.743)
-2.349
(2.429)
3.846**
(1.782)
-0.058
(1.674)
-0.113*
(0.064)
-0.583***
(0.178)
-0.545***
(0.129)
-0.263
(0.175)
0.618***
(0.140)
1.479***
(0.277)
0.248
(0.202)
0.417**
(0.201)
0.22
0.75
0.51
0.13
0.46
0.27
0.30
0.52
0.00
0.98
0.06
0.06
Alpha
0.84
0.72
0.53
0.49
Implied λ (in % a year)
0.4
2.9
2.6
1.0
Half life of convergence
(in years)
173
24
27
69
constant
Note: This table includes the results of regressing equation 25. Dependent variable is
and *** denotes significance level at 1%, 5%, and 10% respectively.
75
Numbers in parentheses are t-statics.*, **,
Texas Tech University, Kolthoom Alkofahi, May 2014
same steady state level of per worker income. This can be seen from the speed of
conditional convergence, where the fastest rate is found for the high income
developing sample (λ=2.9%) and the lowest is found for the low income developing
sample (λ= 1%). ranges from 0.1 % to 2.9 % a year.
The half life convergence column shows that it takes 27 years for the middle
income developing country to reach half of the distance between its initial position and
its steady state. The table also shows high estimates of α for all the samples. One
expects that if a country is away from its steady state, it should devote more share of
output per worker to saving. In turn this would increase the production level and
production growth, consequently, speed up the convergence rate.
In general, one can conclude that, based on cross-sectional analysis, the results
of Table IV.A and Table IV.B confirm finding evidence of conditional convergence
which support the validity of the Solow model, however, the rates of convergence are
very slow. The estimated output shares with respect to capital are still higher than
implied by the model. As discussed before, this upward bias could be corrected
through the inclusion of Foreign Direct Investment.
2. THE CONDITIONAL CONVERGENCE BASED ON FDI
This section studies the effect of initial level of income per worker on the
subsequent growth rates when FDI is incorporated. This study takes the growth
equation of the Solow model and induces FDI as another control variable.
To form a judgment of whether FDI exert any possible positive effect on
economic growth, and if it expedite countries’ convergence rates, equation (25)
requires some necessary modifications. To include FDI into the model, let’s first
define some important abbreviations.
76
Texas Tech University, Kolthoom Alkofahi, May 2014
Let
be the fraction of income devoted to investment in physical capital, and
be
the fraction of income devoted to investment in FDI, hence, equation (25) is modified
to
(26
Thus, the growth rate of income per worker is a function of the log of initial
income per worker, average capital share as a percentage of GDP, log of average
working age population, and the log of average FDI as a percentage of GDP. Equation
(26) is estimated using OLS cross country approach. The main objective of this
section is to assess whether adding FDI into the growth regression attributes more to
the inverse relationship between initial income per worker and economic growth,
thence, the speed of convergence. Another objective shall be comparing the quality of
the parameters in both, MRW human capital augmentation, and our analysis with FDI
augmentation. For example, up to this stage, the estimated share of output with respect
to physical capital is considerably high, would such inclusion of FDI lower the
estimates of α in the way human capital accumulation did for MRW analysis? Finally,
form a final judgment of, in the phase of OLS cross-sectional approach, of the
legitimacy of the Solow model or its extension in explaining income differences
across countries.
DISCUSSION OF RESULTS
CASE I: SAMPLES OF MRW
Table V.A reported the results of regressing equation (26) that aim to test how
far the analysis will change in the presence of FDI. The modification made to equation
77
Texas Tech University, Kolthoom Alkofahi, May 2014
(25) partially affects the outcomes that gave less weight of the true FDI impact. Before
we expand on this discussion, several issues should be outlined.
The first panel of the table gives results of estimation in unrestricted form, while the
second panel contains results from the estimation after imposing the restriction.
Unfortunately, the results are unexpectedly very disappointing. FDI negatively affects
economic growth of the Non-oil sample, and positively for the other two samples.
However, these coefficients enter the regression insignificantly. The estimated
coefficients for the initial level of income almost unchanged, which means that the
chances of convergence in all three samples remain steady in
the presence of FDI. This may indicate that, based on cross-sectional analysis,
countries in the samples do not benefit from FDI. In another word, FDI failed helping
countries to utilize their resources efficiently. Which is literary untrue.
The restricted model is rejected for the Non-Oil and Intermediate samples.
However, steady implied values of λ, and similar estimates values of α are obtained
from the restricted regression.
Before jumping to the conclusion, we need to discuss if choosing different
groups of countries will be able to absorb the FDI efficiently. For this we use the
samples that we constructed.
78
Texas Tech University, Kolthoom Alkofahi, May 2014
Table V.A: Single cross-section results of conditional convergence Augmented Solow model.
Developing and sub developing samples
Sample
# of countries
constant
Implied λ (in % a year)
Half life convergence (in
years)
Non-Oil
Intermediate
(84)
(74)
Unrestricted regression
***
-4.255
-4.193***
(0.830)
(0.720)
***
-0.110
-0.128***
(0.045)
(0.044)
0.483***
0.547***
(0.122)
(0.110)
***
-1.556
-1.681 ***
(0.352)
(0.315)
-0.020
0.053
(0.056)
(0.054)
0.29
0.94
0.39
0.32
0.4
0.5
173
139
79
OECD
(24)
1.643
(1.748)
-0.467***
(0.103)
0.454
(0.274)
-0.856
(0.495)
0.035
(0.063)
0.61
0.20
2.1
33
Texas Tech University, Kolthoom Alkofahi, May 2014
Table V.A. Continued
Sample
# of countries
Constant
Alpha
beta
Implied λ (in % a year)
Half life convergence (in
years)
Non-Oil
(84)
Intermediate
(74)
Restricted regression
***
-2.385
-2.392***
(0.577)
(0.539)
-0.069
-0.114***
(0.045)
(0.044)
0.605***
0.657***
(0.121)
(0.114)
-0.038
0.010
(0.058)
(0.057)
0.22
0.93
0.41
0.34
0.000
0.000
0.95
0.84
-0.060
0.013
0.2
0.4
346
173
Note: This table includes the results of regressing equation 26. ; Dependent variable is
**, and *** denotes significance level at 1%, 5%, and 10% respectively
80
OECD
(24)
1.755
(1.735)
-0.457***
(0.100)
0.571**
(0.239)
0.038
(0.063)
0.61
0.20
0.38
0.54
0.04
2.0
35
Numbers in parentheses are t-statics.*,
Texas Tech University, Kolthoom Alkofahi, May 2014
CASE II: DEVELOPING COUNTRIES AND SUBSAMPLES
The Developing samples are classified based on income classification.
Table V.B reveals unexpected results regarding the contribution of FDI to economic
growth and, consequently, the speed of convergence. Even though evidence of
convergence holds significantly for high and middle income developing samples,
countries in the samples are converge at about the same rates of convergence.
Surprisingly, FDI coefficient is negative and insignificant for the Developing, middle,
and low income samples, and positive but insignificant for the high income
developing sample. The inclusion of FDI hurts the performance of the low income
economy especially that the speed of convergence is getting smaller. In the presence
of
FDI, countries in the low income sample converge at a rate of 0.1% a year compared
to 0.4% a year. (Syed T. Jawaid, Syed Raza,2012) illusterate that the negative impact
of FDI may be a
result of different structural factors. The introduction of new technologies requires the
existence of skilled labor in the host country, which are capable and trained of using
those technologies. If the supply of labor is short in host country than it leads to
negative impact on production and economic growth.
Another possible reason of negative impact may include the imperfect
competitive market. Entrance of foreign companies in the imperfect competitive
markets may lead to reduce market share of domestic producers. Capabilities of scale
economies also suffer in domestic producers because of lost of market share, which
has a negative impact on productivity. On the other hand, data supports the existence
of convergence at a higher rate for the high income developing sample. The
coefficient of initial level of income per worker is (-0.661) in the presence of FDI
compared to (-0.583) without FDI, which means that FDI contributes to economic
growth and countries in the sample are now closer to their steady states. Countries in
this sample grow at rate 3.6% a year and take 19 years to reach half way of the
respective steady states compared to 2.9% a year with 24 years to reach the same
81
Texas Tech University, Kolthoom Alkofahi, May 2014
steady states. This finding refutes the conclusion of Weil (2008) where he addressed
that the average growth rate of GDP per capita over the period 1965-2000 in a closed
economy was around 1.5%, and 3% for open economies. The estimated elasticity with
respect to capital for the intermediate sample in the presence of FDI is no 33%
compared to 53% .However; the implied estimates of (α) for the rest of the samples
remain very high even though FDI is considered as another factor of output.
82
Texas Tech University, Kolthoom Alkofahi, May 2014
Table V.B: Single cross-section results of conditional convergence Augmented Solow model.
Developing and sub developing samples
Sample
# of countries
constant
Implied λ (in % a year)
Half life convergence (in
years)
Developing all
High income
(64)
(14)
Unrestricted regression
***
-6.101
-1.929
(1.407)
(2.951)
Middle income
(20)
Low income
(30)
0.712
(2.321)
-9.006**
(3.888)
-0.081
(0.060)
-0.661***
(0.201)
-0.468***
(0.129)
-0.027
(0.182)
0.490***
(0.139)
1.367***
(0.325)
0.149
(0.202)
0.329
(0.194)
-2.187***
(0.527)
-1.711**
(0.692)
-1.312**
(0.587)
-3.331**
(1.227)
-0.077
(0.071)
0.116
(0.107)
-0.005
(0.116)
-0.187*
(0.107)
0.32
0.43
0.3
231
0.73
0.28
3.6
19
0.56
0.28
2.1
33
0.27
0.48
0.1
693
83
Texas Tech University, Kolthoom Alkofahi, May 2014
Table V.B. Continued
Sample
# of countries
Constant
Alpha
Beta
Implied λ (in % a year)
Half life convergence (in
years)
Developing all
(64)
-2.178***
(0.744)
High income
(14)
Restricted regression
-1.525
(2.527)
Middle income
(20)
Low income
(30)
3.843**
(1.835)
-0.183
(1.677)
-0.102
(0.065)
-0.663***
(0.192)
-0.542***
(0.134)
-0.227
(0.178)
0.638***
(0.141)
1.409***
(0.282)
0.254
(0.211)
0.455**
(0.204)
-0.078
(0.077)
0.22
0.46
0.00
0.95
-0.11
0.4
173
0.107
(0.099)
0.75
0.27
0.76
0.65
0.05
3.6
19
-0.022
(0.126)
0.48
0.30
0.07
0.33
-0.02
2.6
26
-0.116
(0.113)
0.13
0.52
0.02
0.80
-0.18
0.9
77
Note: This table includes the results of regressing equation 26. ; dependent variable is
denotes significance level at 1%, 5%, and 10% respectively
84
Numbers in parentheses are t-statics.*, **, and ***
Texas Tech University, Kolthoom Alkofahi, May 2014
COMMENT- CROSS SECTIONAL FRAMWORK
In general, one can summarize the results obtained by the cross sectional analysis:
1. Constructing new revised and extended data do not produce different results
than those of MRW; constructing samples in a way to be structurally more
homogenous is not the way to solve the problems arisen using the OLS crosssectional approach.
2.
Most of the new constructed samples fail to support some of the Solow
model’s implications.
3.
Including FDI as another determinant of economic growth has positive but
insignificant effect on level of income and economic growth for samples
similar to MRW. However, the effect of FDI is correlated with the way
samples are constructed.
4. The issue of convergence- that stands as a pillar to backup the validity of the
Solow model against the endogenous growth theory- holds for all samples.
However, inclusion FDI could not accelerate the speed of convergence. This
finding contradicts the finding of cross sectional advocates of the contribution
of FDI to economic growth, such as (Syed T Jawaid, Syed A. Raza).
85
Texas Tech University, Kolthoom Alkofahi, May 2014
CHAPTER VII
PANEL DATA ANALYSIS
The choice of appropriate estimation technique is important for obtaining
robust estimates. Some of the existing literature on growth uses cross sectional
approach to estimate the impact of FDI on economic growth, however, many panel
estimations advocates claim that this formulation ignores the country specific aspects
of the data that may be correlated with explanatory variables, causing omitted variable
bias. This point is reinforced by Islam (1995), CEL (1996), Lee, esaran, and Smith
(1997), BHT (2001), Nawaz (2011) and many other researchers. The main
assumption in cross section approach is the strict exogeneity of explanatory variables
that may be violated in many cases. Although this problem can be tackled using
instrumental variables technique, it is very difficult to find valid instrument. The
problem of omitted variables can also be tackled by employing a panel approach
where cross-sectional units are surveyed over time. The time invariant or constant
heterogeneity (that is associated with political situation, geography, and other country
specific factors) might affect the quality of parameter estimates if not properly
addressed. This problem can also be removed using the panel estimation technique
based on pooled, fixed effects, or random effects approaches.
The pooled ordinary least squares (POLS) estimation is the simplest panel
methodology which is more suitable for static cross-sectional data analysis. However,
this method fails to account for the time-series dimension of data since it puts all
observations together into a “pool” and creates deficiency; it fails to account for the
unobserved country-specific (fixed) effects that cause an omitted variable bias, which
then is picked up by the error term, along with the correlation between some of the
independent variables and country-specific effects. In most cases, the pooled OLS
86
Texas Tech University, Kolthoom Alkofahi, May 2014
approach is unlikely to be adequate, but it provides a baseline for comparison with
more complex estimator.
Furthermore, a fixed effects model allows each cross-section unit (country) to
have its own intercept. The intercept varies across countries, but it is time-invariant.
The random effects model, on the other hand, calculates the common intercept as
being a mean value of all cross-section units and the error term is the deviation of each
intercept from the mean. The random effects model also assumes that unobservable
individual effects are random variables and are distributed independently of the
regressors. A Hausman test can be run to determine which model is more suitable for
this study. It is assumed that if the cross-section specific error component and the
regressors are uncorrelated, the random effects model is preferred; otherwise, the fixed
effects model is more appropriate.
CHOICE OF ESTIMATOR
Which panel method should one use, pooled, fixed effects or random effects
estimators?
According to this paper, I will use the pooled OLS as a first panel estimation to
control for the unobserved country specific effects. The reason behind using pooled
OLS is to make comparison in conformity with other researches. On the other hand,
the problem of heterogeneity cannot be solved using pooled OLS. If the problem of
heterogeneity arise, then it is necessary to choose if heterogeneity is modeled as either
Random effects of fixed effects. The advantage of employing pooled OLS model is
that this kind of estimation offers additional three tests that allow comparing between
pooled OLS against the alternatives; fixed and random effects models.
87
Texas Tech University, Kolthoom Alkofahi, May 2014
TYPE OF TESTS
Once the pooled OLS is being employed, three diagnostic tests are available; these
tests can be performed to decide whether or not the Pooled OLS is more adequate than
Fixed or random effects models.
1. Joint significance of differing group means
: Pooled OLS is adequate
: Fixed effects is adequate
A low p-value counts against the null hypothesis in favor of the
alternative.
2. Breusch-Pagan test static
: Pooled OLS is adequate
: Random effects is adequate
A low p-value counts against the null hypothesis in favor of the
alternative
3. Hausman test static
: Random effects is adequate
: Fixed effects is adequate
A low p-value counts against the null hypothesis in favor of the
alternative
In order to proceed with the analysis, some data modifications are necessary
for the panel estimation to be feasible. In the cross sectional approach, we used the
average values of the independent variables over the whole period. In order to switch
from a single cross section to a panel framework, we divide the total period into
several shorter spans. Considering the period 1980-2010, I opt for six non overlapping
intervals of five-year time spans. For example, investment, working age population,
and FDI are averaged over five-year time span instead of the full interval. Therefore,
data for saving, working age population, and FDI that are averaged over the period
1980-1985 are available at 1985. Considering this setup, the data for these variables
88
Texas Tech University, Kolthoom Alkofahi, May 2014
are available at 1985, 1990, 1995, 2000, 2005, and 2010. The initial levels of income
per worker are available at these stage points as well as at 1980.With this setup, the
error terms are now five calendar years apart and hence may be thought to be less
influenced by business cycle fluctuations and less likely to be serially correlated than
they would be in a yearly data setup.
Recall, one of this paper’s objectives is to study the validity of the Solow and
Augmented Solow models. However, the main topic of this paper is to study the effect
of FDI on the level of income per worker and its effect on the growth rate of Income
per worker. In order to see how much the results of this paper differ from those of
MRW because of utilizing panel data technique, I will reformulate equation (18’) and
(22) to be adequate for such analysis. invoke equation (18’) and (22)
MRW relied on a crucial assumption regarding the term [
assumed that, since technology is a public good to be equally shared, (
]. They
is the same
for all countries and for a cross-section regression, t is just a fixed number which
render the term
in the equation to be constant and equal for all countries, and hence
it is legitimate to just drop it from the equation. However, this is not true for the
term
. MRW noted that this term reflects technology as well as resource
endowments, climate, institutions, and so on. Therefore, this term might be different
across countries. MRW postulated that
constant for country (i) and
, where the term ( ) is
is an error or disturbance term specific to country (i )
in period t. the new formulation of equation (18’) becomes
89
Texas Tech University, Kolthoom Alkofahi, May 2014
Another way the panel estimation is different than cross country estimation is
the decomposition of the error term. While the cross sectional and pooled OLS
estimations consider the error term to be specifically to country (i), the panel
estimation decomposes the error term based on fixed effect or random effect
estimations. For example, in the cross sectional and pooled OLS models, the error
term specific to country (i) at time (t). When testing for the validity of the Solow or
augmented Solow model and their implications, equations (18.a) and (22.a) are the
right equations to estimate. However, for the fixed effects model, the error term
the following decomposition
, where
has
are country specific- time
invariant component that are treated as fixed parameters. The country specific effect
captures the existence of other determinants of a country’s steady state that are not
already controlled for by the equation,
climates, and so on. Finally, (
may reflect differences in technology, tastes,
) is the observation specific error. Substituting in
yielding
And this reduces to
…………….. (a.1
Substituting the above formulation into (18’) and (22) yields
(22.b)
90
Texas Tech University, Kolthoom Alkofahi, May 2014
To estimate the equation using fixed effects model, the above equations are the
right equations to employ.
On the other hand, if the Husman test fails to reject the null hypothesis,
random effects should be used. The error term takes the following
decomposition
. Unlike the fixed effect model,
s are not treated as
fixed parameter, but as a random drawings from a given probability distribution.
Using this specification, equation (18’) and (22) reduce to
(22.c
We may note that, the above specifications were based on approximation
around the steady state. Also, it may be noted that in the single cross-section
regression,
and
are assumed to be constant for the entire period. Such an
approximation is more realistic over shorter periods of time.
ESTIMATION RESULTS
This section focus on re-estimating the level of income per worker upon the
exogenous variables using pooled OLS and fixed effects models. The Hausman test is
performed; the results of the test produce very low estimates of p-values (0.000) for all
the samples in the study. This suggests that the null hypothesis is rejected and that a
Fixed Effects model produces better coefficient’s estimates. Therefore, all regressions
are estimated using a fixed effect specification. However, in attempt to see if Random
effects produce different results than the fixed effects estimation, I tested the
regressions for all the samples using random effects and found that both estimations
produce results that are literary very similar. Moreover, additional panel estimation
91
Texas Tech University, Kolthoom Alkofahi, May 2014
(Pooled OLS) is performed for the sake of comparison with cross-sectional OLS, and
with pooled OLS that are presented by Islam (1995).
Equations (18.a, 18.b, 22.a and 22.b) are estimated for all the samples in the
study, where the results are reported separately. Each table includes the estimate of the
Solow model and the augmented Solow model using Pooled OLS and Fixed Effects
panel estimations in both unrestricted and restricted forms. The discussion of the
results of each table is explained individually in details. The results are shown in
tables below.
Non-Oil Sample:
We want to find out quantitatively how far the results are different by applying
the panel data approach than those obtained using cross sectional OLS estimation. We
accomplish this by comparing the results of Table I.A column one with column (1)
and (2) in table below.
We find that the impact of using panel data approach is striking. According to
pooled OLS and fixed effects results, the estimate of the coefficients for saving and
working age population are statistically significant, very close in magnitude, and
opposite in sign. Moreover, the restricted model is not rejected at 95% and 98%
significance levels respectively, which indicate that countries are at the end of the
period at their steady state level of per worker income. However, the estimates of
capital share of income per workers create dispute; even though the pooled OLS
estimate of capital share of income per worker is 57% compared to cross sectional
estimated 67% , both estimate are considered higher than implied by the national
accounts information. Furthermore, the fixed effects model finds a very low estimate
(α= 16%), this finding corresponds to CEL (1996) where they found implausibly low
value of α of 10%, accordingly they reject the Solow model. Of For these reasons,
CEL concluded that the Solow model is not adequate for the Non-Oil sample.
92
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VI.A: panel regression analysis, Non-Oil Sample
Model
Estimation
Column #
Constant
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Unrestricted regression
*
***
1.135
8.337
(0.624)
(0.181)
***
***
1.236
0.234
(0.100)
(0.034)
***
*
-1.704
-0.093
(0.213)
(0.050)
0.30
1.105
Constant
Alpha
Beta
0.98
0.200
Restricted regression
1.867
8.234***
(0.501)
(0.177)
***
1.326
0.187***
(0.089)
(0.032)
***
0.30
1.108
0.05
0.57
0.98
0.200
0.02
0.16
Augmented Solow
Pooled OLS
Fixed Effects
(3)
(4)
***
3.928
(0.365)
***
1.004
(0.108)
***
-0.600
(0.058)
***
0.087
(0.033)
0.35
1.05
4.459 ***
(0.342)
0.652***
(0.096)
0.078***
(0.034)
0.33
1.067
0.00
0.38
0.05
***
8.566
(0.181)
***
0.162
(0.038)
***
-0.087
(0.048)
***
0.043
(0.008)
0.98
0.19
8.486***
(0.178)
0.122***
(0.033)
0.043***
(0.008)
0.98
0.190
0.03
0.11
0.04
Note: the output represent pooled OLS, Fixed effect for textbook and augmented Solow model.. Dependent variable is
Numbers in parentheses are t-statics.*, **, and *** denotes significance level at 1%, 5%, and 10% respectively
93
.
Texas Tech University, Kolthoom Alkofahi, May 2014
Proceeding to estimate the augmented version of the model, we find very
encouraging results. First, unlike the results of cross-sectional approach, the effect of
FDI is found to be not only positive, but also highly statistically significant for both
pooled OLS and fixed effects estimates.
The coefficient of FDI is estimated to be (0.043). It implies that, a one unit
increase in the net inflows of FDI, leads to 4.3% increase in the level of income per
worker. The adjusted
for the fixed effects model is substantially higher (0.98); the
model successfully explains the differences in income per worker across countries
using these three control variables, including FDI.
The inclusion of FDI could
substantially lower the estimate of α (38% compared to 57%), however, the restricted
model is rejected for the pooled OLS even though the estimates for α is very close to
the implied value. The fixed effects estimate produces a share of income per worker
that is away below 33%. Both panel estimates demonstrate that the share of income
per worker with respect to FDI is approximately 5%.
Intermediate Sample
It seems from the results that dividing the growth period into five-year spans
has significant effect. The coefficients for
and
are significant
and has the predicted sign and magnitude, especially when the Fixed effects panel
estimation is conducted. Even though the restricted regression is significantly not
rejected, the results of pooled OLS produce estimate for α that is too high relative to
the standard assessment (α=57%), and this estimate is not different than the one
obtained from the cross-sectional analysis. The fixed effects, on the other hand,
produce a very low estimate of α (12%) that is way below the standard assessment.
Testing the impact of FDI on economic growth has the expected outcomes; FDI
positively and significantly affect the level of income per worker. A one unit increases
in FDI leads 5.3% increase in the level of income per worker; countries with high
94
Texas Tech University, Kolthoom Alkofahi, May 2014
share of FDI tend to have higher per worker income. Obviously, the effect of FDI on
the level of income for the Intermediate sample is larger than that of the Non-oil
sample. The coefficient of FDI for the intermediate sample is 0.053 compared to 0.43
for the Non-oil sample. The results also show that the coefficients sum to zero for both
panel estimations. The restricted model is not rejected for both estimates,
unfortunately, as in the Non-oil sample, the pooled OLS over estimated capital share’s
of income, while the fixed effects finds low estimate of α. Bothe pooled and the OLS
techniques show that the share of income per worker with respect to FDI ranges from
6% - 7%.
95
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VI.B: Results of panel regression analysis, Intermediate Sample.
Model
Estimation
Column #
Constant
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Unrestricted regression
*
***
1.038
8.715
(0.622)
(0.191)
***
***
1.264
0.151
(0.107)
(0.043)
***
***
-1.769
-0.128
(0.204)
(0.049)
0.33
0.99
Constant
Alpha
Beta
0.98
0.18
Restricted regression
4.815
8.709***
(0.534)
(0.191)
1.377***
0.141***
(0.094)
(0.033)
***
0.32
1.000
0.03
0.58
0.98
0.190
0.72
0.12
Augmented Solow
Pooled OLS
Fixed Effects
(3)
(4)
*
***
1.090
(0.636)
***
1.216
(0.114)
***
-1.796
(0.202)
***
0.143
(0.033)
0.35
0.98
8.877
(0.186)
***
0.110
(0.042)
***
-0.116
(0.046)
***
0.053
(0.008)
0.98
0.171
1.663***
(0.561)
1.318***
(0.101)
0.146***
(0.033)
0.34
0.98
0.06
0.54
0.07
8.867***
(0.186)
0.090***
(0.034)
0.052***
(0.007))
0.98
0.171
0.43
0.09
0.06
Note: the output represent pooled OLS, Fixed effect for textbook and augmented Solow model.. Dependent variable is
Numbers in parentheses are t-statics.*, **, and *** denotes significance level at 1%, 5%, and 10% respectively
96
.
Texas Tech University, Kolthoom Alkofahi, May 2014
The OECD Sample
Unlike the other samples, when cross country approach is conducted, data for
the OECD sample supports the Solow model’s implications regarding the sign of the
coefficients and the estimate of α. However, dividing the full period into six subperiods leads the coefficients to be opposite in sign and equal in magnitude. The
estimates of α matches the national account estimate (α= 1/3) for pooled OLS
approach. Even though fixed effects panel approaches produce similar value of α, the
restricted model reject the null hypothesis.
Noticeable effect of FDI is also observed, FDI positively and significantly
affects the level of income per worker for the OECD countries. According to the
results, among the samples that are constructed similar to MRW, the largest affect of
FDI is viewed for the OECD sample. The coefficient of FDI is (0.118) compared to
(0.043, and 0.053) for the Non-oil and Intermediate samples respectively. We can
interpret this result as: a one unit increase in FDI increases the level of income per
worker by 11.8%.
The restricted model is not rejected for pooled OLS, but is rejected for the
fixed effects estimation. The implied value of α that matches the national accounts
information is found by the pooled OLS approach. The estimate of output per worker
share with respect to FDI ( ) is equal to 6%. The fixed effect uncovers a value of α
that equals to 18% which is far from being close to the standard assessment. Data for
the OECD sample shows that
ranges from 6% to 9% and this confront to what is
found by using the cross sectional framework. Finally, one can notice that the
inclusion of FDI into the equation increases the fit of the regression;
compared to 82%.
97
is now 90%
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VI.C: Results of panel regression analysis, OECD Sample.
Model
Estimation
Column #
Constant
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Unrestricted regression
***
8.028
7.253***
(0.685)
(0.500)
***
0.370
0.851***
(0.149)
(0.132)
***
-0.585
-0.321***
(0.169)
(0.095)
0.10
0.37
0.84
0.16
98
Augmented Solow
Pooled OLS
Fixed Effects
(3)
(4)
7.474***
(0.678)
0.458***
(0.147)
-0.679***
(0.165)
0.088***
(0.025)
0.16
0.36
8.494***
(0.390)
0.525***
(0.103)
-0.236***
(0.071)
0.118***
(0.012)
0.91
0.118
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VI.C. Continued
Model
Estimation
Column #
Constant
Alpha
Beta
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Restricted regression
***
8.075
7.859***
(0.682)
(0.485)
***
0.463
0.500***
(0.115)
(0.033)
0.10
0.37
0.32
0.32
0.82
0.17
0.00
0.33
Augmented Solow
Pooled OLS
Fixed Effects
(3)
(4)
7.488***
(0.676)
0.517***
(0.112)
0.090***
(0.025)
0.18
0.36
0.54
0.32
0.06
8.998***
(0.378)
0.246***
(0.066)
0.123***
(0.067)
0.90
0.123
0.00
0.18
0.09
Note: the output represent pooled OLS, Fixed effect for textbook and augmented Solow model. Dependent variable is
Numbers in parentheses are t-statics.*, **, and *** denotes significance level at 1%, 5%, and 10% respectively
99
.
Texas Tech University, Kolthoom Alkofahi, May 2014
Sample of Developing Countries
It seems so far that dividing the full period into 6 sub-periods has remarkable
effects. Similar to previous tables, the estimates using the Pooled OLS and Fixed
effects approaches produce lower and significant estimate of
and
.
Unlike the cross sectional analysis, including FDI as another factor of production has
significant effects when both panel estimated are employed. FDI coefficient is
positive, significant, and a one unit increases in the net inflows of FDI leads to a 0.04
increase in the level of income per worker.
The inclusion o FDI in the pooled regression reduces the estimate of α to 46%
compared to 59% in the cross section framework, but as expected, α is higher than one
third. According to the panel fixed effect outcomes, the estimated coefficients of the
regressors matches the Solow model’s implication, the coefficients are opposite in
sign and almost equal in magnitude. The restricted model is not rejected at 90%
significance level. FDI lowers the estimate of α from 15% to 12%, however, these
estimates are considered very low. Furthermore, the share of income per worker that is
devoted to FDI activities is estimated in both models by 4%.
100
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VI.D: Results of panel regression analysis, Developing countries.
Model
Estimation
Column #
Constant
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Unrestricted regression
***
5.156
7.906***
(0.624)
(0.198)
***
0.933
0.205***
(0.090)
(0.040)
***
-0.347
-0.100*
(0.214)
(0.057)
Augmented Solow
Pooled OLS
Fixed Effects
(3)
(4)
4.987***
(0.636)
7.945***
(0.193)
0.925***
(0.098)
0.149***
(0.053)
-0.416***
(0.218)
-0.151***
(0.046)
0.040***
(0.009)
0.97
0.19
0.22
0.96
0.089***
(0.034)
0.23
0.94
0.21
0.92
101
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VI.D. Continued
Constant
Restricted regression
4.078
7.812***
(0.465)
(0.190)
***
0.853
0.171***
(0.085)
(0.035)
***
0.21
0.95
0.01
0.46
0.96
0.22
0.10
0.15
3.960***
(0.489)
7.917***
(0.187)
0.834***
(0.091)
0.136***
(0.036)
0.083***
(0.034)
0.22
0.93
0.01
0.43
0.04
0.040***
(0.009)
0.97
0.19
0.54
0.12
0.04
Note: the output represent pooled OLS, Fixed effect for textbook and augmented Solow model.. Dependent variable is
Numbers in parentheses are t-statics.*, **, and *** denotes significance level at 1%, 5%, and 10% respectively
102
.
Texas Tech University, Kolthoom Alkofahi, May 2014
Sample of High Income Developing Countries
The unrestricted regression of pooled OLS estimate that is used to test the
Solow growth model yields consistent estimates. The restricted form generates value
of output share with respect to capital that is higher than one third (59%).
Assessing whether conducting panel estimation would result in more
reasonable estimates of α is totally feasible. The pooled OLS generates coefficients
that are lower in absolute values than those generated in Table I.B. the restricted
model is not rejected at very high significance level and produce value of α (41%) that
is lower than 59%. Hence dividing the full interval into short subinterval lower the
estimate of α, however, this estimate is not even closed to one third. Adding FDI into
the regression is of a great deal; it decrease the estimate of capital share’s to
reasonable value similar to the estimate implied by the national accounts (α=35). On
the other hand, the evaluated share of output with respect to FDI is 6%. This means
that, high income developing countries devote 6% of its income per worker for such
activities carried by FDI.
Switching from pooled OLS to fixed effects technique increased the fit of the
regression (
) from 0.29 to 0.75 with lower standard error of the regression. It also
generates insignificant coefficients of the saving rate but highly significant for
working age population’s coefficient.
Lastly, it lowers the estimate of α to more acceptable value (0.24%) compared to 0.41
of the pooled OLS regression, where restricted model is not rejected at 97%
significance level.
It seems that the validity of the textbook Solow model depends on the way the
sample is constructed; this can be concluded from the results obtained for this sample.
Adding FDI to the regression could slightly increase the fit of the regression. Akin to
the previous tables, FDI positively and significant at 99% significance level. A one
103
Texas Tech University, Kolthoom Alkofahi, May 2014
unit increases in FDI increase the level of income per worker by 9.1%. Such inclusion
substantially decreases the estimate of α where , the elasticity of income per worker
with respect to FDI, is found to equal 8%.
104
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VI.E: Results of panel regression analysis, High income developing countries.
Model
Estimation
Column #
Constant
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Unrestricted regression
***
6.415
8.145***
(0.713)
(0.601)
Augmented Solow
Pooled OLS
Fixed Effects
(3)
(4)
6.483***
(0.704)
8.529***
(0.609)
0.05
(0.147)
-0.509***
(0.185)
0.091***
(0.028)
0.78
0.21
0.767***
(0.135)
-0.473**
(0.214)
0.106
(0.141)
-0.602***
(0.164)
0.29
0.75
0.666***
(0.134)
-0.545***
(0.205)
0.106***
(0.031)
0.37
0.37
0.22
0.35
105
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VI.E. Continued
Constant
Restricted regression
6.118
8.220***
(0.669)
(0.618)
***
0.685
0.32***
(0.116)
(0.107)
***
0.29
0.37
0.24
0.41
0.73
0.23
0.03
0.24
6.274***
(0.668)
0.60***
(0.118)
0.110***
(0.031)
0.37
0.35
0.35
0.35
0.06
Note: the output represent pooled OLS, Fixed effect for textbook and augmented Solow model. Dependent
variable is
. Numbers in parentheses are t-statics.*, **, and *** denotes significance level at 1%, 5%,
and 10% respectively
106
8.532***
(0.618)
0.211*
(0.111)
0.100***
(0.028)
0.77
0.21
0.10
0.16
0.08
Texas Tech University, Kolthoom Alkofahi, May 2014
Middle Income Developing Sample
Akin to the cross-country OLS, the estimates for the log of saving rate are
measured insignificantly using both panel approaches, for both Solow and augmented
Solow models. Switching from pooled OLS to fixed effects increased the fit of the
regression numerously. Even though the restricted model of pooled OLS estimate is
not rejected at 90% significance level, the estimates of capital share using both
techniques produce outrageously low estimate of α.
On the other hand, FDI enters the regression positively, and a one unit
increases in FDI leads to 3% increase in the level of income per worker. The implied
value of α and
are fund to be very low of (0.08 and 0.04) respectively.
107
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VI.E: Results of panel regression analysis, Middle income developing countries.
Model
Estimation
Column #
Constant
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Unrestricted regression
***
8.388
7.796***
(0.666)
(0.441)
Augmented Solow
Pooled OLS
Fixed Effects
(3)
(4)
8.656***
(0.671)
8.074***
(0.460)
0.02
(0.080)
-0.427***
(0.142)
0.030*
(0.015)
0.83
0.16
0.048
(0.099)
-0.273
(0.226)
0.033
(0.040)
-0.521***
(0.134)
0.00
0.83
0.040
(0.098)
-0.171
(0.229)
0.062*
(0.030)
0.02
0.39
0.16
0.39
108
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VI.E. Continued
Restricted regression
Constant
8.770 ***
(0.521)
0.083
(0.091)
8.347***
(0.424)
0.158
(0.075)
0.81
8775***
(0.513)
0.050
(0.091)
0.064**
(0.030)
0.03
8.562***
(0.420)
0.098
(0.076)
0.041***
(0.015)
0.82
0.21
0.95
0.17
0.38
0.16
0.36
0.00
0.78
0.02
0.08
0.14
0.05
0.08
0.06
0.04
Note: the output represent pooled OLS, Fixed effect for textbook and augmented Solow model.. Dependent variable is
.Numbers in parentheses are t-statics.*, **, and *** denotes significance level at 1%, 5%, and 10% respectively.
109
Texas Tech University, Kolthoom Alkofahi, May 2014
Low Income Developing Countries
Table below represents results of regressing the textbook Solow model and the
augmented version using pooled OLS and fixed effect panel approaches. When the
cross-country framework is employed, data for low income developing countries
perfectly fits the Solow model. However, based on the results below, it is clear that
dividing the full period into shorter spans doesn’t meet the implications of the Solow
model regarding the coefficients. Moreover, the restricted model is rejected and the
values of capital shares are way below one third. For these reasons, the Solow model
for this sample is unambiguously rejected. Testing for the augmented Solow model,
we can see that the coefficient of FDI is now positive compared to the cross sectional
analysis but is insignificant.
The restricted model is not rejected but produces very low estimate of α and
of
(0.12, and0.02) respectively. The direct message one can comprehend is that the low
income countries are largely labor intensive agricultural economies. As the working
age population increases, more workers are devoted to produce more output.
110
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VI.F: Results of panel regression analysis, Low income developing countries.
Model
Estimation
Column #
Constant
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Unrestricted regression
***
7.702
7.285***
(0.492)
(0.234)
Augmented Solow
Pooled OLS
Fixed Effects
(3)
(4)
7.528***
(0.509)
7.156***
(0.219)
0.210***
(0.013)
-0.043
(0.061)
0.020
(0.028)
0.90
0.19
0.239***
(0.074)
0.206
(0.167)
0.232***
(0.048)
0.046
(0.068)
0.06
0.86
0.268***
(0.082)
0.163
(0.174)
0.020
(0.028)
0.06
0.60
0.23
0.60
111
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VI.F. Continued
Model
Estimation
Column #
Constant
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Restricted regression
***
6.874
7.025***
(0.374)
(0.229)
**
0.179
0.150***
(0.071)
(0.043)
0.02
0.62
0.01
0.15
0.86
0.23
0.00
0.13
Augmented Solow
Pooled OLS
Fixed Effects
(3)
(4)
6.757***
(0.401)
0.200**
(0.078)
0.017
(0.029)
0.03
0.60
0.02
0.16
0.02
7.023***
(0.618)
0.142***
(0.044)
0.024*
(0.012)
0.90
0.20
0.10
0.12
0.02
Note: the output represent pooled OLS, Fixed effect for textbook and augmented Solow model.. Dependent variable is
. Numbers in parentheses are t-statics.*, **, and *** denotes significance level at 10%, 5%, and 1% respectively
112
Texas Tech University, Kolthoom Alkofahi, May 2014
COMMENTS-PANEL DATA ANALYSIS
Several issues need to be highlighted:
1. Based on the Hausman test, the analysis of fixed effects estimation is more
adequate than using pooled OLS estimation. However, pooled OLS estimator
is used for cross country comparisons.
2. Dividing the full interval into shorter periods of five years span leads to
striking results. The coefficients of the equations become highly significant,
and even support the Solow’s implications regarding the sign and magnitude
for most of the samples.
3.
Despite concludes (1) and (2), the restricted models are not rejected but yields
in estimates of capital’s share of income per worker that are lower than one
third. The pooled OLS generates upward biased estimates of α, whereas the
fixed effects approach downward biases the corresponding estimates.
4. Elasticity of output with respect to capital appears to be very low for the
middle income and low income developing samples, using both panel
techniques.
5. It is clear by now that the results are somehow different based on the way
samples are constructed; grouping countries does matter.
6. For these reasons, we may reject the validity of the Solow models for all the
samples except for OECD and high income developing countries. However,
instead of rejecting such well known model, we can accept the model’s validity
in explaining Income differences across countries, and reconcile the finding of
α by generalizing that: countries are no longer devoting 33% of its income to
capital. This value is lower nowadays and literary depends on the development
level of each sample (or country). Especially, that there are different types of
capital that are available for households to choose from to invest in.
7. FDI affects the level of income positively, and for most of the samples the
effect is highly significant.
113
Texas Tech University, Kolthoom Alkofahi, May 2014
8. Incorporating FDI as another factor of input could lower the estimate of α to
match the value implied by the national accounts. This can be seen when
pooled OLS is using Non-Oil and high income developing samples.
9. The share of output per worker with respect to FDI is not uniformed across the
samples. This is represented by the estimated values of . The lowest value
appears in the low income developing counties where =2%, this means that,
countries of low income developing samples devotes only 2% of its income per
worker for FDI activities. Whereas OECD and high income developing
countries devote around 9% of its income per worker for such activities. We
carry similar argument when the augmented model is employed.
10. Finally, and most importantly, the main objective of this literature is to find if
FDI exert any positive effects. As we have found, FDI contributes to the level
of income, and that countries at the steady states differ based on the saving
rate, working age population, and the volume of FDI that has entered each
country. Therefore, we suggest that each economy should work harder to
attract most of FDI, and be the candidate for such capital flow. For example, as
an effort to attract FDI and spur economic growth, many developing countries
have established investment agencies and have introduced policies that include
fiscal and financial incentives.
114
Texas Tech University, Kolthoom Alkofahi, May 2014
CHAPTER VIII
TESTING FOR CONDITIONAL CONVERGENCE PREDICTIONS
OF THE SOLOW AND AUGMENTED SOLOW MODELS
In the growth literature, three empirical results have surfaced: (1) absence of
absolute convergence among countries in the larger sample, (2) Slow conditional
convergence among countries in the larger sample, and (3) absolute of faster
conditional convergence among similar subgroups of countries of the larger sample.
As we have mentioned earlier, it was the absence of absolute convergence that
triggered the development if new endogenous theories of growth. However, the
concept of conditional convergence is thought to reinstate the Solow growth model.
The preceding panel estimations built on the assumptions that countries are at
or near their own steady states per worker income. It is possible, though, to utilize a
more general framework that examines the predictions of the Solow model for
behavior of income per worker out of steady states. Such framework allows estimation
of the effect of various explanatory variables on per-worker growth rates as well as the
speed at which actual income per worker reaches the steady state level of income per
worker. Hence, one of the objectives of this section is finding evidence of convergence
using both panel estimations; pooled OLS and fixed effects.
Accordingly, we need to find out how far the results are different when
applying panel data estimations. Finding evidence of convergence is thought to be one
way to support the validity of the textbook or augmented Solow models. Moreover,
the main objective falls in finding substantiation on whether FDI positively affect the
growth rate of income per worker, and if rates of convergence are manipulated by the
presence of FDI. One should take into consideration that countries may experience
different growth patterns depending on their development level.
115
Texas Tech University, Kolthoom Alkofahi, May 2014
Finding the speed of converge in panel estimation is somehow different than
the approach conducted using cross-sectional framework. This difference is made to
account for some issues that may arise using the cross-sectional technique such as
mitted variable bias and /or heterogeneity across samples. Following Islam’s (1995),
the general form of the standard growth regression model is estimated:
Where
includes the control variables, and p is the number of variables included in
the regression. P may take the value of 2 or 3 based on whether or not FDI is
incorporated.
Before we expand on this equation, it is important at this stage to illustrate how
to derive conditional convergence using panel estimation for the reader to
comprehend.
Let’s define some variables of interest:
the steady state level of income per effective worker.
the actual level of income per effective worker at time t.
Approximating around the steady, the speed of convergence is given by
Where
The speed of convergence is the speed at which actual income is reaching its steady
state level of income measured in percentage per year.
The above equation implies that
116
Texas Tech University, Kolthoom Alkofahi, May 2014
Where
is income per effective worker at some initial point of time, and
. In the panel estimation,
takes different value than in single cross sectional
approach. It refers to the difference between the end and the beginning of each period,
hence, for the sake of our study,
Substituting for
is just equal to 5.
(equation 12) yields:
.b
The above equation has been formulated in terms of income per effective
worker. We may, therefore, reformulate the equation in terms of income per worker.
Note that income per effective worker is
Where:
is the income per worker, substituting the above equation into (24) we
get
For simplicity, we can rewrite the above equation as below:
Where:
117
Texas Tech University, Kolthoom Alkofahi, May 2014
=
,
=
=
,
=
, and
=
is the general disturbance term which includes the unobservable country specific
effect (
,
is a time specific effect represented by year dummies, and the
transitory error term (
that varies across countries and time periods, and has mean
equals to zero.
Subtracting
Where:
=
from both sides yields
, and all other terms have similar interpretations as before.
Similarly, finding conditional convergence in the presence of FDI takes similar
arguments. The general form of the equation is:
……… (29
Where:
Equations (26) and (27) represent the conditional convergence equations that
capture the dynamics toward the steady state. These equations focus on the ability to
reduce the income gap between the current state and steady state.
118
Texas Tech University, Kolthoom Alkofahi, May 2014
In order to answer the questions of interests, we regress equations (26), and (27) using
pooled OLS and fixed effects estimators for all the samples. The results are displayed
in Table VII.1 through Table VII.7. The tables include the results acquired from
regressing the restricted and the unrestricted model. Based on the Hausman test, the
fixed effects estimator is more adequate approach than other techniques.
In this context, we need to reevaluate the panel estimations that we have
obtained in Table VI.A through Table VI.F to confirm the validity of the textbook or
the augmented Solow models. First of all, the findings of a faster conditional
convergence (even without taking account of FDI) pack up the validity of the crosscountry implications of the Solow model. Second, based on the discussion of panel
estimation, the steady state levels of income per worker differ across countries not
only because of the control variables, but also because of differences in term
.
Whether or not differences in this term are important solely depend on the results
obtained. For example, if the results acquired using panel framework are not
remarkably different than cross-sectional analysis, then less weights is giving to this
term. Indeed, our results show that differences in
term plays an important role in
understanding income differences across countries. We then turn to the question of
what happens when FDI is brought into the growth regression after controlling
for
. We expect that using panel estimation technique, FDI plays a significant
role in enhancing economic growth. However, this role varies depending on the
development of each sample.
To proceed with the analysis, we first discuss whether or not dividing the full
periods into shorter spans and considering the growth process into shorter consecutive
intervals has any significant effect on the issue of convergence. For this reason, we
first compare the results of the restricted and the unrestricted conditional convergence
obtained by pooled OLS and cross country framework. We then analyze how
accounting for individual country effects would change the results. This can be done
by comparing the outcomes between the pooled OLS and fixed effects estimation.
119
Texas Tech University, Kolthoom Alkofahi, May 2014
Similarly, we compare the results when FDI is induced into the model. We hope to
find that, not only FDI positively affect the growth process, but also accelerate the rate
at which countries are converging, and reduce the implied estimate of output share
with respect to physical capital.
DISCUSSION OF RESULTS: SAMPLES OF MRW
CASE I: CONVERGENCE IN SOLOW MODEL
A. POOLED OLD VERSUS CROSS-COUNTRY APPROACH
The results obtained for Non-oil, Intermediate, and OECD samples underline
the existence of inverse relationship between the initial level of income per worker and
subsequent growth rates. This finding stands as evidence of conditional convergence.
The coefficients of
and
obtained from the cross sectional and pooled OLS approaches are
similar in the spirit that they have the predicted sign, and very close in magnitude.
However, high estimates of are found which contradicts the Solow model’s
implications. Moreover, using pooled OLS approach leads to lower the estimates of
the coefficients of
and the rate of conditional convergence. According to the
restricted regression, when cross-sectional approach is employed, the estimates for λ
are (0.3%, 0.4%, and 2%), and the corresponding pooled OLS estimates are (0.10%,
0.24%, and 1.89%) for the Non-oil, Intermediate, and OECD samples respectively.
120
Texas Tech University, Kolthoom Alkofahi, May 2014
B. PANEL FIXED EFFECTS VERSUS POOLED OLS APPROACHES
By looking at the first and second columns of each table, one can conclude that
controlling for the individual country effects and measurement errors leads to
substantial change in the results. In the fixed effects approach, the coefficients of
initial level of income are highly significant, and increased substantially (in absolute
value). Since the restricted regression is not rejected at very high significance levels,
we aim at comparing the results of the restricted model only. Therefore, for the Nonoil, Intermediate, and OECD samples, these coefficients equal to (-0.313, -0.221, and 0.211) compared to (-0.007,-0.012, and -.090) when pooled OLS approach is
employed. Consequently, the implied value of λ was the highest for the Non-oil
sample and the lowest for OECD sample. This finding confirms both types of
convergence; absolute convergence, where poor countries tend to grow faster than rich
countries eventually reaching to the same level of income per worker, and in term of
conditional convergence, where countries converge to their respective steady states.
The fixed effect approach generates values of λ (7.5%, 5.0%, and 4.74%) that
are higher than the corresponding pooled OLS estimates. Significant change is also
observed in the implied values of α; the estimated values of α is (0.22, 0.26, and 0.40),
these values truly contrast with the corresponding pooled OLS estimates of (0.93,
0.88, and 0.53). Finally, employing the fixed effects estimator fits the model bitter
than pooled OLS estimator, this can be seen from the implied values of the adjusted
(0.24, 0.24, and 0.29) compared to the corresponding pooled OLS estimates of
(0.05, 0.09, and 17).
The above findings imply that, correcting for the individual country specific
effect is highly informative. For the purpose of testing the validity of the Solow
model, we find some evidences that allow us to consider the model as adequate in
explaining income differences across countries. First, the coefficients of the saving
rate and the working age population are opposite in sign and very close in magnitude.
121
Texas Tech University, Kolthoom Alkofahi, May 2014
Second, the estimated values of α are somehow close to one third. Finally, finding
evidence of a fast rate of convergence is though as another evidence to support the
Solow growth model.
In sum, we conclude that, constructing samples similar to those of MRW with
new revised and extended data, considering out of steady states behavior, and
estimating the model using fixed effects estimator lead to remarkable change in the
results; we obtain: higher estimates of the initial level of income per worker, much
higher rates of convergence, more empirically plausible estimates of the elasticity of
output with respect to capital, and better fit of the Solow model. For these reasons, we
are able to reject the endogenous growth model in favor of the e Solow growth model.
CASE II: CONVERGENCE IN THE AUGMENTED SOLOW MODEL
Having seen the impact of inclusion the individual country effects on growth
regression results, we now reevaluate the question that is considered the heart of our
analysis: what happens when FDI is brought into the model, when panel framework
analysis is conducted. We intend, in this section, to discuss several issues. First, Akin
to the discussion above, we compare the regression of the restricted model before and
after dividing the full periods into shorter intervals. Second, we compare the results
between estimates produced by fixed effects and pooled OLS estimations. Finally, we
investigate if such augmentation affects the values of the structural parameters, the
overall the performance of the regression, therefore, the validity of the augmented
Solow model.
A. POOLED OLS VERSUS CROSS SECTIONAL APPROACH
Referring to Tables IV.B, VII.1, VII.2, and VII.3; first thing to recap is that,
using pooled OLS, the coefficient of
becomes positive and highly significant
for the Non-oil sample, and statistically significant for the Intermediate sample.
122
Texas Tech University, Kolthoom Alkofahi, May 2014
However, for the OECD sample, the coefficient of
statistically insignificant. The coefficients for
becomes lower but
are negative, significant, and
become lower in absolute values. The estimate of α is substantially lower for the Nonoil and Intermediate samples (0.74, and 0.75) compared to (0.93, and 0.84), where the
value of α is remarkably unchanged for the OECD sample (α=0.53). On the other
hand, the highest estimate of output elasticity with respect to FDI is found for the
Non-oil sample (0.18) and the lowest for the OECD sample (0.2), based on these
results, poor countries devote more income saving to FDI activities than rich
countries. The pooled estimations resulted in slightly lower estimates of λ for the
Intermediate and OECD samples compared to cross sectional results.
In short, dividing the full sample into shorter periods and adding FDI into the
regression lead to considerable changes, what we are mostly interested in is that, FDI
exert positive effect on output growth for MRW samples, and that it generate lower
estimates of α, however, these estimates are still with very large values.
B. POOLED OLS VERSUS PANEL FIXED EFFECT ESTIMATIONS
We now turn to discuss if controlling for the correlated individual country
effect leads to different estimates. We scan columns (3) and (4) of each table and
compare the results obtained. Briefly speaking, it is with the fixed effects estimation
that we observe: negative and higher estimates of the initial level of income, more
plausible estimates of the saving rates and working age population, higher elasticity
with respect to FDI, faster speed of convergence, better fit of the model, and lower
estimates of α. We can conclude that, incorporating FDI is better estimated once we
control for correlated individual country effect, this is also predicted once we test for
the best choice of estimator using joint significance test.
123
Texas Tech University, Kolthoom Alkofahi, May 2014
COMMENTS: THE ROLE OF FDI ON ECONOMIC GRWOTH USING FIXED
EFFECTS APPROACH AND SAMPLES OF MRW
The core if this section is to shed some light on changes that affects the economy
once FDI is incorporated into the model. Based on our analysis above, we accepted the
Solow model’s implications; we turn now to test whether the data fits the augmented
Solow model’s implication. For this reason, we compare the results obtained after
incorporating FDI into the model (column 4) with those of column (2). We can
conclude that adding FDI into the growth process leads to remarkable results:

First thing to notice is that, the coefficients of
are negative, statistically
significant, and even higher in absolute value for all the samples.

Estimates of the control variables coefficient’s matches, in the spirit, to Solow
model’s implication; they are opposite in sign and very close in magnitude.

The log of FDI inters the regression positively and significantly for all the
samples. Unexpectedly, these measurements are almost the same for all the
samples. The corresponding estimates of FDI for the Non-Oil, Intermediate, and
OECD samples are (0.035, 0.030, and 0.035). One can interpret these results as, a
1% increase in FDI leads to increase economic growth .

More than 32% of variations in income per worker across courtiers are explained,
not only by variations in the control variables and the term (
), but also by
variations in the net inflows of FDI. This is manifested by the values of
adjusted

.
Adding FDI into the regression speed up the rate at which countries are
converging. λ is found to be the highest for the Non-oil sample, and the lowest for
the Intermediate sample. These estimates are ( 8.40%, 6.13%, and 7.25%)
compared to the corresponding values without FDI (7.50%, 5.0%, and 4.74%).
These estimates are unarguably much higher than the values generally accepted by
growth advocates; between 2% and 4%.
124
Texas Tech University, Kolthoom Alkofahi, May 2014

Finding evidence of a faster rate of conditional convergence somehow lends more
validity to the Solow model. Our analysis in this regard finds that FDI provides
evidence of a faster rate of convergence; it also discovers that the relative increase
in λ was the highest for the OECD sample and the lowest for the Non-oil sample
(12%, 27%, and 53%). This may assure the validity of the Solow model and that
the countries of OECD sample are the big beneficiaries for such FDI activities.

The estimated values of α after including FDI into the regression for the Non-oil
and Intermediate samples are (0.10, and 0.15) compared to (0.22, and 0.26), these
estimates are strikingly very low. Therefore, if the purpose of the study is to find
lower estimates of α, then augmenting the model is not necessary, since the
estimates that produced by the fixed effects estimation are acceptable. However,
for the OECD sample, the inclusion of FDI reduces the estimates of α to 0.25. We
fail to reject the augmented Solow model for the OECD sample.

The share of income with respect to FDI ( ) is estimated as (0.08, 0.09, and 0.08).
This means that, countries devote almost 8% of its income per worker for FDI
activities. This is our remarkable finding that, based on my knowledge, is not
estimated in any other literature.

After all, we conclude, based on the fixed effects estimates that, FDI positively
and significantly affect the growth rate of income per worker. FDI accelerate
economic growth and convergence for the Non-oil, Intermediate and the OECD
samples. For this reason, we suggests that, poor countries must reconsider more
policies to attract more foreign investors at home, for it is now clear that FDI
helps countries to converge to their steady states and to reduce the income
inequality across countries.
125
Texas Tech University, Kolthoom Alkofahi, May 2014
TABLE VII.1: Test for conditional convergence: Non-oil sample
Model
Estimation
Column #
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Unrestricted regression
-0.253***
2.539***
(0.090 )
(0.302)
-0.004
-0.310***
(0.006)
(0.032)
***
0.078
0.140***
(0.016)
(0.026)
-0.040
-0.017
(0.032)
(0.035)
Augmented Solow
Pooled OLS
Fixed Effects
(3)
(4)
0.05
0.26
-0.238***
(0.091)
-0.006
(0.006)
0.073***
(0.017)
-0.045
(0.032)
0.020***
(0.005)
0.08
0.16
0.14
0.16
0.13
Implied λ (in % a year)
0.08
7.42
0.12
8.3
Half life of convergence
(in years)
864.7
9.3
575.9
8.3
Constant
126
2.978***
(0.297)
-0.340***
(0.032)
0.088***
(0.027)
-0.021
(0.034)
0.035***
(0.005)
0.34
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.1. Continued
Model
Estimation
Column #
Constant
Alpha
Beta
Implied λ (in % a year)
Half life of convergence
(in years)
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Restricted regression
-0.311 ***
2.443***
(0.073)
(0.306)
-0.005
-0.313***
(0.006)
(0.033)
***
0.071
0.087***
(0.015)
(0.023)
0.05
0.16
0.27
0.93
0.24
0.14
0.00
0.22
0.10
691
7.50
9.2
Augmented Solow
Pooled OLS
Fixed Effects
(3)
(4)
-0.304***
(0.076)
-0.007
(006)
0.066***
(0.016)
0.016***
(0.005)
0.06
0.16
0.19
0.74
0.18
0.14
493
Note: This table includes the results of regressing equations 28 and 29. Dependent variable is
parentheses are t-statics.*, **, and *** denotes significance level at 10%, 5%, and 1% respectively
127
2.911***
(0.301)
-0.343***
(0.032)
0.038
(0.023)
0.035***
(0.005)
0.32
0.13
0.00
0.10
0.08
8.40
8.3
. Numbers in
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.2: Test for conditional convergence: Intermediate Sample
Model
Textbook Solow
Estimation
Column #
Pooled OLS
Fixed Effects
(1)
(2)
Unrestricted regression
-0.384***
1.692***
(0.078)
(0.311)
constant
Implied λ (in % a year)
Half life of convergence
(in years)
Augmented Solow
Pooled OLS
(3)
Fixed Effects
(4)
-0.382***
(0.080)
2.194***
(0.310)
-0.012**
(0.006)
-0.220***
(0.032)
-0.016***
(0.006)
-0.263***
(0.032)
0.084***
(0.015)
-0.112***
(0.027)
0.097***
(0.027)
-0.055*
(0.031)
0.09
0.12
0.24
287
0.23
0.11
4.97
14
0.087***
(0.016)
-0.119***
(0.027)
0.014***
(0.004)
0.11
0.12
0.32
215
0.071***
(0.027)
-0.050*
(0.029)
0.030***
(0.005)
0.32
0.10
6.10
11.4
128
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.2. Continued
Model
Textbook Solow
Estimation
Column #
Pooled OLS
Fixed Effects
(1)
(2)
Restricted regression
***
-0.346
1.690***
(0.067)
(0.313)
Constant
-0.012**
(0.006)
0.090***
(0.014)
Alpha
Beta
Implied λ (in % a year)
Half life of convergence
(in years)
-0.221***
(0.032)
0.079***
(0.021)
0.09
0.12
0.35
0.88
0.24
0.11
0.29
0.26
0.24
287
5.00
13.9
Augmented Solow
Pooled OLS
(3)
Fixed Effects
(4)
-0.359***
(0.071)
2.186***
(0.309)
-0.015***
(0.006)
0.091***
(0.014)
-0.264***
(0.032)
0.049**
(0.021)
0.014***
(0.004)
0.030***
(0.005)
0.11
0.12
0.53
0.75
0.12
0.30
229.3
Note: This table includes the results of regressing equations 28 and 29. Dependent variable is
parentheses are t-statics.*, **, and *** denotes significance level at 10%, 5%, and 1% respectively
129
0.32
0.11
0.17
0.15
0.09
6.13
11.3
. Numbers in
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.3: Test for conditional convergence: OECD Sample
Model
Estimation
constant
Implied λ (in % a year)
Half life of convergence
(in years)
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Unrestricted regression
0.447**
1.492***
(0.216)
(0.391)
***
-0.091
-0.204***
(0.018)
(0.042)
**
0.071
0.115
(0.035)
(0.076)
-0.138***
-0.150***
(0.040)
(0.048)
Augmented Solow
Pooled OLS
Fixed Effects
(3)
(4)
0.17
0.28
0.410*
(0.216)
-0.090***
(0.019)
0.079**
(0.035)
-0.139***
(0.040)
0.003
(0.006)
0.16
0.09
1.91
36.3
0.08
4.56
15.2
0.08
1.89
36.8
130
2.558***
(0.508)
-0.303***
(0.052)
0.112
(0.072)
-0.150***
(0.045)
0.035***
(0.010)
0.36
0.074
7.22
9.6
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.3. Continued
Model
Estimation
Constant
Alpha
Beta
Implied λ (in % a year)
Half life of convergence
(in years)
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Restricted regression
0.449 **
1.501***
(0.217)
(0.389)
***
-0.090
-0.211***
(0.018)
(0.039)
***
0.100
0.141***
(0.027)
(0.043)
0.17
0.09
0.18
0.53
0.29
0.08
0.68
0.40
1.89
36.7
4.74
14.6
Augmented Solow
Pooled OLS
Fixed Effects
(3)
(4)
0.411*
(0.216)
-0.090***
(0.019)
0.104***
(0.028)
0.004
(0.006)
0.16
0.08
0.26
0.53
0.02
1.89
36.7
Note: This table includes the results of regressing equations 28 and 29. Dependent variable is
parentheses are t-statics.*, **, and *** denotes significance level at 1%, 5%, and 10% respectively
131
2.561***
(0.500)
-0.304***
(0.048)
0.114**
(0.041)
0.035***
(0.010)
0.36
0.074
0.98
0.25
0.08
7.25
9.56
. Numbers in
Texas Tech University, Kolthoom Alkofahi, May 2014
DISCUSSION OF RESULTS: DEVELOPING AND SUBDEVELOPING
SAMPLES
One of our contributions in this study is to find out if the results are correlated
with the way samples were grouped. In other world, do we accept the Solow model for
each economy? Does the Solow model applicable based on the level of development
of the country? For this reason, we follow the analysis above using developing
countries samples and the subsamples that we constructed; we see whether or not the
fixed effects approach generates better results than those produced by the pooled OLS
approach. In another study, we test the validity of the Solow model and its
augmentation. We see what changes are made by incorporating FDI into the growth
regression.
CASE I: CONVERGENCE IN SOLOW MODEL
A. POOLED OLS VERSUS CROSS-COUNTRY APPROACH
This part uses the results of estimating conditional convergence of the Solow
model using pooled OLS and compares them to those of cross-sectional estimates
reported in Table IV.B. The samples to consider in the comparison are the developing
and sub-developing samples, whose results are interpreted in Table VII.4 through
Table VII.7. The results of pooled OLS confirm the evidence of an inverse
relationship between the initial level of income per worker and subsequent growth
rates, however, the coefficients are estimated with lower values. The coefficients of
saving rate and growth rate of working age population are similar to previous tables in
the sense that, they have the predicted opposite sign, and very close in magnitude.
However, when comparing the results before and after subdividing the full intervals,
we find that pooled OLS approach leads to a lower values of the initial level of
income(
) and the rate of conditional convergence , and leads to higher estimates
for α.
132
Texas Tech University, Kolthoom Alkofahi, May 2014
B. PANEL FIXED EFFECTS VERSUS POOLED OLS APPROACHES
Switching from pooled OLS estimation to fixed effects estimation, where the
individual country effects and measurement errors are being controlled for, leads to
substantial change in the results. In the fixed effects approach, the coefficients of
initial level of income for the developing, (high income, middle income, and low
Income) developing samples are highly significant and are equal to ( -0.322, -0.220, 0.267, and -0.407). These estimates are larger, in absolute value than the corresponding
estimates of pooled OLS (-0.013, -0.085, -0.097, and -0.042), all of which are found
statistically significant at 10% significance level. Coefficient of the initial level of
income is found the largest (in absolute value) for the low income developing sample,
and is the lowest for high income developing sample. This finding matches the
prediction of convergence in the Solow model where it hypothesized that, countries
with lower initial income per worker tends to grow more rapidly. This is correctly
interpreted when we look at the implied speed of convergence and the half life
convergence. The implied values of λ for the unrestricted model are (7.77%, 5.0%,
6.21%, and 10.45%). As we observe, low income developing countries converge to its
steady state at the highest rate of 10.45% each year, therefore, it takes only 6.6 years
to reach half way to its steady state level of income per worker. The results also
confirm that, the high income developing country is converging at the lowest rate of
5%, which means that, the developing countries may recover in term of income per
worker and eliminate the gap of income inequality possibly reaching the same steady
state.
However, when testing the validity of the Solow model, we see that, it is only
for the low income developing sample that the coefficient of working age population
is found significantly positive. This might be due to a measurement error, more
possibly that it is positive since the economy of low income countries is based on the
agriculture sectors, hence more workers available increase the production level. For
133
Texas Tech University, Kolthoom Alkofahi, May 2014
this finding, we conclude that the Solow model’s implications are not applicable for
the low income developing countries.
Another significant change one might notice is the value of
samples. The values of
for each of the
for the Developing country and sub-samples are estimated
as (0.29, 0.33, 0.45, and 0.33) compared to pooled OLS estimations of (0.06, 0.33,
0.31, and 0.07).
The restricted model, on the other hand, is not rejected at 95% significance
level for all the samples. The estimates of the initial level of income are negative and
measured significantly. Consequently, the implied values of λ are (8.10%, 4.44%,
5.49%, and 10.48%) for the developing and sub developing samples. These estimates
are definitely larger than the corresponding pooled OLS estimates of (0.22%, 1.86%,
2.11%, and 0.62%). We can see that the estimates of λ are the highest for the low
income developing sample and the lowest for the high income developing sample.
Again, this matches the hypothesis of convergence where poor countries converge at a
faster rate, eventually, catching up with the rich countries in term of income per capita.
Another remarkable change is a lower estimate of α obtained by the model using fixed
effects approach. α is equal to (0.24, 0.47, 0.46, and 0.17) compared to pooled OLS
estimates of (0.88, 0.74, 0.52, and 0.62). As we can see, the estimates of α for the
high and middle income developing samples are considered higher than the predicted
value by the model, and the value is considered very low for the low income
developing sample.
Based on the discussion above, the fixed effects panel approach produce more
desirable results, compared to pooled OLS estimation. The Solow model is not
rejected for the developing country sample. Because of large estimate of α, this study
entails more investigation of whether the augmented Solow model is more reliable.
134
Texas Tech University, Kolthoom Alkofahi, May 2014
COMMNTS: THE ROLE OF FDI ON ECONOMIC GRWOTH USING FIXED
EFFECTS APPROACH, AND DEVELOPING AND SUBDEVELOPING
SAMPLES
The Solow model is rejected for the low income developing sample; because
of the positive coefficient of working age population growth rate. However, would we
reject the augmented Solow model as well? is it true that, when multinational firms
bring in new investment to the host low income developing countries, their economy
become more capital reliance countries? This will be discussed when we include FDI
into the growth regression.
Similar argument is carried on to high and middle income developing samples,
where the Solow model didn’t perfectly in, because of high estimates of α. Hence,
what changes in the results could be made when FDI is included in the model?
To answer these questions, we run the growth regression incorporating FDI as
another factor of input, and compare the results of column 2 and 4 of each table. We
can conclude that adding FDI into the growth process leads to remarkable results:

That, the coefficients of
for the augmented Solow model are (-0.341, -
0.246, -0.270, and -0.454) these are unarguably negative, statistically significant,
and even higher in absolute value than the corresponding results of the Solow
model for the developing and sub-developing samples of (-0.322, -0.220, -0.267,
and -0.407). The results emphasize that FDI is important factor in enhancing
economic convergence where the relative increase in the coefficients is the 12%
for the high and the low income developing samples.

The coefficients of the control variable are now opposite in sign and similar in
magnitude for all the samples including the low income developing sample. This is
important since the effect of the working age population growth was positive for
the low income developing sample. With the FDI in place, this effect becomes
negative and inconformity with the augmented Solow model’s implications.
135
Texas Tech University, Kolthoom Alkofahi, May 2014

As expected, FDI positively and significant effect economic growth for all the
samples. Unlike the samples of MRW, this coefficient is not the same across the
samples. For the developing and the sub-developing samples, this coefficient is
(0.039, 072, 0.027, and 0.260) respectively. Among the developing samples, the
high income developing sample is largely affected by the inflows of FDI; a one
percentage change in the net inflows of FDI yields a 0.72% increase in the growth
rate of income per worker. The middle and the low income developing samples are
similarly affected by the net inflows of FDI; a one percentage change in the FDI
leads to 0.026% change in the growth rate of income per worker.

Including FDI considerably increased the fit of the regression for the developing
sample, high income developing countries, and low income developing counties.
The value of
after including FDI is (0.38, 0.50, 0.49, and 0.40) and the value
before including FDI is (0.29, 0.33, 0.45, and 0.33). This means that variation in
income per worker across courtiers are explained, not only by variations in the
control variables and the term (
), but also by variations in the net inflows of
FDI.

Adding FDI into the regression accelerate the rate of convergence. The values of λ
after and before the inclusion of FDI are found as (8.62%, 5.34% 5.86%, and
12.47%) compared to (8.10%, 4.44%, 5.495%, and 10.865). The speed of
convergence is found the highest for low income developing sample confirming
that poorer countries are converging at a faster rate than richer countries. These
estimates are unarguably much higher than the values generally accepted by
growth advocates.

Finding evidence of a faster rate of conditional convergence somehow lends more
validity to the Solow model. Our analysis in this regard finds that FDI provides
evidence of a faster rate of convergence; it also discovers that the relative increase
in λ was the highest for the OECD sample and the lowest for the Non-oil sample
(12%, 27%, and 53%). This may assure the validity of the Solow model and that
the countries of OECD sample are the big beneficiaries for such FDI activities.
136
Texas Tech University, Kolthoom Alkofahi, May 2014

The estimated values of α after including FDI into the regression for the samples
are (0.16, 0.22, 0.36, and 0.13) compared to (0.24, 0.47, 0.46, and 0.17). We can
see that the augmented Solow model is now valid for the high and middle income
developing countries.

CEL (1996) rejected the Solow model and its augmentation for very low estimate
of α and very fast speed of convergence. One way to reconcile the findings of very
low estimate of α is to claim that, the implied value of α is no longer one third,
since different types of capital are available in each economy; such as, FDI, human
capital, health… etc, hence, more varieties of capital means less share of output
with respect to physical capital.

The share of income with respect to FDI ( ) is estimated as (0.09, 0.19, 0.08, and
0.05). Obviously, the high income developing countries are devoting 19% of its
income per worker for FDI activities. On average, the developing countries devote
9% of its income per worker for FDI activities, and this finding is in conformity
with those obtained by the MRW sample. This is our remarkable finding that,
based on my knowledge, is not estimated in any other literature.

After all, we conclude, based on the fixed effects estimates that, FDI positively
and significantly affect the growth rate of income per worker. FDI accelerate
economic growth and convergence for the developing and sub-developing
samples.
137
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.4: Test for conditional convergence: Developing Sample
Model
Estimation
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Unrestricted regression
-0.208
2.415***
(0.127 )
(0.344)
-0.013
-0.322***
(0.009)
(0.039)
***
0.096
0.158***
(0.018)
(0.029)
-0.030
-0.014
(0.040)
(0.041)
Textbook Solow
Pooled OLS
Fixed Effects
(3)
(4)
0.06
0.29
-0.232***
(0.117)
-0.025***
(0.009)
0.094***
(0.018)
-0.083**
(0.034)
0.026***
(0.006)
0.13
0.18
0.15
0.16
0.13
Implied λ (in % a year)
0.26
7.77
0.51
8.34
Half life of convergence
(in years)
264.9
8.9
136.9
8.31
Constant
138
2.577***
(0.318)
-0.341***
(0.036)
0.113***
(0.028)
-0.040
(0.037)
0.039***
(0.006)
0.38
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.4. Continued
Model
Estimation
Constant
Alpha
Beta
Implied λ (in % a year)
Half life of convergence
(in years)
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Restricted regression
-0.337 ***
2.357***
(0.096)
(0.350)
-0.011
-0.333***
(0.009)
(0.039)
***
0.0855
0.105***
(0.017)
(0.026)
0.05
0.18
0.12
0.88
0.26
0.16
0.09
0.24
0.22
313.3
8.10
8.6
Textbook Solow
Pooled OLS
Fixed Effects
(3)
(4)
-0.301***
(0.090)
-0.024***
(0.009)
0.087***
(0.017)
0.026***
(0.006)
0.13
0.16
0.36
0.64
0.19
0.49
142.7
Note: This table includes the results of regressing equations 28 and 29. Dependent variable is
parentheses are t-statics.*, **, and *** denotes significance level at 1%, 5%, and 10% respectively
139
2.565***
(0.321)
-0.350***
(0.036)
0.076***
(0.025)
0.040***
(0.007)
0.37
0.13
0.15
0.16
0.09
8.62
8.1
. Numbers in
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.5: Test for conditional convergence: High Income Developing Sample
Model
Estimation
Constant
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Unrestricted regression
-0.556
1.175*
(0.407 )
(0.694 )
**
-0.085
-0.220***
(0.041)
(0.067)
0.234
0.062
(0.056)
(0.082)
-0.282***
-0.338***
(0.080)
(0.098)
0.23
0.14
Augmented Solow
Pooled OLS
Fixed Effects
(3)
(4)
0.33
-0.244
(0.389)
-0.120***
(0.039)
0.211***
(0.055)
-0.315**
(0.076)
0.042***
(0.012)
0.32
1.771***
(0.619)
-0.246***
(0.059)
0.002
(0.078)
-0.272**
(0.086)
0.072***
(0.015)
0.50
0.13
0.13
0.11
Implied λ (in % a year)
1.78
5.0
2.56
5.65
Half life of convergence
(in years)
39.0
14.0
27.1
12.3
140
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.5. Continued
Model
Estimation
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Restricted regression
-0.483
1.031
(0.380)
(0.708)
-0.089
-0.199***
(0.040)
(0.068)
***
0.249
0.177***
(0.047)
(0.063)
Augmented Solow
Pooled OLS
Fixed Effects
(3)
(4)
-0.157
(0.377)
-0.124***
(0.038)
0.231***
(0.047)
0.042***
(0.011)
0.33
0.13
0.49
0.58
0.11
2.65
26.2
1.674***
(0.625)
-0.235***
(0.059)
0.087
(0.060)
0.077***
(0.015)
0.48
0.11
0.10
0.22
0.19
5.34
13.0
Note: This table includes the results of regressing equations 28 and 29. Dependent variable is
parentheses are t-statics.*, **, and *** denotes significance level at 1%, 5%, and 10% respectively
Numbers in
Constant
Alpha
Beta
Implied λ (in % a year)
Half life of convergence
(in years)
0.24
0.14
0.61
0.74
0.30
0.13
0.04
0.47
1.86
37.2
4.44
15.6
141
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.6: Test for conditional convergence, Middle Income Developing Sample
Model
Estimation
Constant
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Unrestricted regression
-0.156
1.025*
(0.288 )
(0.563 )
***
-0.097
-0.267***
(0.024)
(0.054)
**
0.062
0.128**
(0.027)
(0.049)
-0.351***
-0.419***
(0.062)
(0.080)
Augmented Solow
Pooled OLS
Fixed Effects
(3
(4)
1.272**
(0.556)
-0.270***
(0.052)
0.118**
(0.048)
-0.341***
(0.082)
0.027***
(0.009)
0.49
0.31
0.45
-0.030
(0.290)
-0.102***
(0.023)
0.058**
(0.026)
-0.323***
(0.062)
0.022**
(0.009)
0.34
0.11
0.10
0.10
0.09
Implied λ (in % a year)
2.04
6.21
2.15
6.29
Half life of convergence
(in years)
34
11
32
11
142
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.6. Continued
Model
Estimation
Constant
Alpha
Beta
Implied λ (in % a year)
Half life of convergence
(in years)
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Restricted regression
0.359
1.094*
(0.282)
(0.589)
***
-0.100
-0.240***
(0.025)
(0.056)
***
0.106
0.204***
(0.027)
(0.044)
0.20
0.11
0.00
0.52
0.40
0.10
0.05
0.46
2.11
32.9
5.49
12.6
Augmented Solow
Pooled OLS
Fixed Effects
(3
(4)
0.435
(0.271)
-0.108***
(0.025)
0.091***
(0.026)
0.031***
(0.009)
0.27
0.11
0.00
0.39
0.14
2.29
30.3
1.373***
(0.563)
-0.254***
(0.053)
0.161
(0.044)
0.033***
(0.009)
0.48
0.09
0.11
0.36
0.08
5.86
11.8
Note: Results of regressing equations 28 and 29. Dependent variable is
. Numbers in parentheses are tstatics.*, **, and *** denotes significance level at 1%, 5%, and 10% respectively.
143
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.7: Test for conditional convergence, Low Income Developing Sample
Model
Estimation
Constant
Implied λ (in % a year)
Half life of convergence
(in years)
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Unrestricted regression
0.415
3.041***
(0.263 )
(0.468 )
-0.042
-0.407***
(0.026)
(0.060)
**
0.080
0.171***
(0.026)
(0.038)
0.112*
0.120**
(0.026)
(0.054)
Textbook Solow
Pooled OLS
Fixed Effects
(3
(4)
3.256***
(0.444)
-0.454***
(0.057)
0.150***
(0.040)
-0.036
(0.048)
0.026***
(0.010)
0.40
0.07
0.33
0.343
(0.237)
-0.055**
(0.024)
0.087***
(0.025)
0.047
(0.053)
0.021**
(0.009)
0.11
0.21
0.18
0.18
0.15
0.89
10.45
1.1
12.10
80.8
6.6
61.3
5.7
144
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.7
Model
Estimation
Constant
Alpha
Beta
Implied λ (in % a year)
Half life of convergence
(in years)
Textbook Solow
Pooled OLS
Fixed Effects
(1)
(2)
Restricted regression
-0.011
2.841***
(0.231)
(0.500)
-0.033
-0.419***
(0.027)
(0.065)
**
0.053
0.083**
(0.025)
(0.036)
0.02
0.21
0.00
0.62
0.23
0.19
0.06
0. 17
0.67
103.3
10.86
6.4
Textbook Solow
Pooled OLS
Fixed Effects
(3
(4)
0.021
(0.208)
-0.047**
(0.024)
0.061**
(0.024)
0.020**
(0.009)
0.08
0.19
0.01
0.48
0.16
0.96
72.2
Note: This table includes the results of regressing equations 28 and 29. Dependent variable is
parentheses are t-statics.*, **, and *** denotes significance level at 1%, 5%, and 10% respectively
145
3.179***
(0.465)
-0.464***
(0.060)
0.074
(0.035)
0.030***
(0.010)
0.34
0.15
0.13
0.13
0.05
12.47
5.6
. Numbers in
Texas Tech University, Kolthoom Alkofahi, May 2014
STUDY CASE: THE ISSUE OF CONDITIONAL CONVERGENCE IN
ISLAM (1995)
As mentioned in the literature review, Islam (1995) re-estimated the work of
MRW using pooled OLS and the fixed effects approach. Our interest shall fall upon
this work since studying the issue of convergence is the sole of Islam’s work.
Therefore, we aim in this section at comparing the results obtained in table VII.1 with
those of Islam, especially Tables (II, IV, and V).
Before we expand on this comparison, one should note that the results
produced by our analysis are not directly comparable to Islam work; Islam uses the log
of income per capita as the dependent variable while in our analysis, we use the
growth rate of income per worker. We, therefore, look at the implied values of the
initial level of income per worker and (α); we can simply subtract the coefficients of
the initial level of income per capita from one and use the outcome to compare the
results. Moreover, Islam’s work covers 25 consecutive years (1960-1985), by dividing
this full interval into five-year span, his work includes 5 intervals. However, our work
covers 30 consecutive years, which implies that 6 intervals are included in the study.
Table II of Islam produces results of pooled OLS regression estimating the
conditional convergence using MRW samples, Table IV includes results of the
estimation using fixed effects panel estimation, and Table V displays the regression
for the Non-oil sample when the model is augmented with human capital
accumulations. We take advantages of his work and try to show the robustness of
using such new data and samples, whether the speed of convergence is higher, and if
FDI is better than human capital accumulation in enhancing engine income
convergence.
Table VII.1 and Table VII.2 summarizes the results of estimating the restricted
conditional convergence obtained from pooled OLS and fixed effects estimations. The
146
Texas Tech University, Kolthoom Alkofahi, May 2014
top part of the table represents the outcomes acquired by Islam, whereas the lower part
represents our corresponding estimates. For reasons explained earlier, we look at the
implied values of the structural parameters only.
Based on pooled OLS estimation, we found a lower rate of convergence for the
Non-oil and Intermediate samples than Islam, where both results confirm of a very
slow rate of convergence.
The Table also shows that our estimates of (α) for the Non-oil and intermediate
samples are much higher than those estimated by Islam.
However, the fixed effects approach yields considerable changes in the results; the
estimated rate of convergence for the Non-oil sample is higher in our study, but lower
for the intermediate and OECD samples. Unlike our findings, Islam found the rate of
convergence is the largest for the OECD sample, which means that, rich countries are
getting richer and the gap between poor countries and rich countries are getting larger.
Significant change is also observed in the implied values of the elasticity
parameter, α. All of which are thought to be lower than those obtained by pooled OLS
approach and much closer to the generally accepted values. Based on Islam’s results,
the value of α is found larger for the Non-oil and the Intermediate samples, and
smaller for the OECD samples.
In Sum, constructing samples similar to MRE using more revised and extended
data, and the adoption of the panel data approach leads to lower estimates of physical
share of income per worker for the Non-oil and Intermediate samples.
147
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.8: Restricted Conditional convergence using Pooled OLS
Pooled OLS
Islam
This paper
Implied
parameters
λ
α
λ
α
Non-oil
0.0059
0.83
0.0008
0.93
Samples
Intermediate
0.0095
0.77
0.002
0.88
OECD
0.0146
0.62
0.016
0.53
Note: dependent variable
Table VII.9: Restricted Conditional convergence using fixed effects
Fixed effects
Islam
This paper
Implied
parameters
λ
α
λ
α
Non-oil
0.047
0.44
0.063
0.22
Samples
Intermediate
0.046
0.46
0.041
0.26
OECD
0.093
0.21
0.039
0.40
Note: dependent variable
Similar to MRW, Islam investigates the effect of including human capital
accumulation into the model when panel estimation is adopted. We departed from
using human capital accumulation into using FDI instead. Our next goal is then to
compare between the two types of augmentation and discuss if the affect of FDI is
larger than the affect of human capital accumulation.
According to Islam, results based on pooled OLS estimation show that, the
human capital variable does not prove to be significant for all the samples, and it
assumes the wrong sign for the Intermediate sample. The inclusion of human capital
variable lowers the estimate of λ, and considerably increased the value of α.
Consequently, our analysis using pooled OLS regression leads to similar results;
however, the coefficient of FDI is significantly positive for all the samples. This
discussion is summarized in Tables VII.3 through Table VII.5 below.
148
Texas Tech University, Kolthoom Alkofahi, May 2014
Turning to fixed effects estimation, the human capital variable generates
negative but insignificant effect for all the samples. Islam stated that, the implied
speed of convergence and the estimated values of α are broadly similar to the results
without including human capital variable. The implied exponent for the human capital
variable was found negative for all the samples. Our finding, however, leads to results
in a very different direction.
We conclude that, constructing samples similar to those of MRW, with revised
and extended data, and FDI as another factor of input generates better and more
plausible results than the approach of Islam. The inclusion of FDI leads to a larger
effects on economic growth, accelerates the speed of convergence, and lowers the
share of income per worker with respect to capital, unlike the effects produced by the
inclusion of human capital variable. Thus, in spite of the negligible importance that
human capital accumulation played in the growth context of Islam, we think that
countries should put heavy prominence on the importance of FDI and not to ignore the
generally accepted perception of the importance role that human capital accumulation
plays in increasing the development level of each country.
Table VII.10: Restricted Conditional convergence (augmented model)
Non-Oil sample
ISLAM w/ H.K
parameter
Pooled OLS
Fixed effects
λ
0.0069
0.0375
α
0.80
0.05
0.001
0.52
-0.20
0.07
0.74
0.18
0.10
0.08
λ
This paper w/FDI
α
Note: dependent variable
149
Texas Tech University, Kolthoom Alkofahi, May 2014
Table VII.11: Restricted Conditional convergence (augmented model)
Intermediate sample
parameter
λ
Pooled OLS
0.0079
Fixed effects
0.0444
ISLAM w/ H.K
α
This paper w/FDI
λ
α
0.79
-0.008
0.003
0.75
0.12
0.50
-0.0069
0.051
0.15
0.09
Note: dependent variable
Table VII.12: Restricted Conditional convergence (augmented model)
OECD sample
ISLAM w/ H.K
parameter
Pooled OLS
λ
0.0162
0.091
α
0.601
0.017
0.016
0.207
-0.045
0.062
0.53
0.02
0.25
0.08
λ
This paper w/FDI
Fixed effects
α
Note: dependent variable
150
Texas Tech University, Kolthoom Alkofahi, May 2014
CHAPTER IX
FDI AND THE ISSUE OF CONVERGENCE UNDER THE GMM
ESTIMATION
It turned out that many authors criticized the existing cross-country empirical
research on economic growth showing that the statistical assumptions underlying such
work are violated. They postulated and emphasized on some sources of inconsistency
in existing cross-country empirical work on economic growth: correlated individual
effects and endogenous explanatory variables. The cross-country approach failed to
tackle these two major flaws, therefore, the correlation between lagged dependent
variables and the unobserved residual that does not disappear with time averaging, is
precisely the reason why we prefer the panel data over the cross-sectional approach
when analyzing growth effects.
The method of fixed effects is designed to control for the unobserved country
specific time invariant effects in the data, however, advocates of dynamic panel
estimation criticized this panel method claiming that, it does not address the problem
of endogeneity. According to them, even though fixed effects correct for the possible
correlation between lagged dependent variables and the unobserved residual, it do so
by taking deviations from time averaged sample means, where the dependent variable
is stripped of its long-run variation that make it inappropriate for studying a dynamic
concept.
The first-differenced generalized method of moments (1st-diff GMM) estimator
of Arellano and Bond (1991) that is applied to dynamic panel data models is one of
the renowned procedures to address these problems. The basic idea is to write the
Solow model equation as a dynamic panel data model, take first differences to remove
unobserved time-invariant country-specific effects, and then instrument the right hand
side variables in the first-differenced equations using levels of the series lagging two
151
Texas Tech University, Kolthoom Alkofahi, May 2014
periods or more, taking into consideration that the time-varying disturbances in the
original level equations are not serially correlated.
When studying growth literature, we can see that the 1 st-diff GMM has
important advantages over simple cross-section regressions and other estimation
methods for dynamic panel data models: (i) the estimates will no longer be biased by
any omitted variables that are constant over time (unobserved country specific or fixed
effects). In conditional convergence regressions, this avoids the problem raised by the
omission of initial efficiency. (ii) The 1 st-diff GMM estimator uses instrumental
variables in the regression that allows parameters to be estimated consistently in
models which include endogenous right-hand-side variables, such as investment rates.
It also allows consistent estimation even in the presence of measurement error.
The second type of GMM estimator is the system GMM (SYS-GMM) by Arellano
and Bover (1995) and Blundell and Bond (1998). Blundell and Bond system GMM
uses both lagged level observations as instruments for differenced variables and
lagged differenced observations as instruments for level variables. These instruments
are valid under restrictions on the initial conditions to obtain moment conditions that
remain informative even for persistent series.
As in the 1st-diff GMM estimator, the SYS-GMM has one set of instruments to
deal with endogeneity of regressors and another set to deal with the correlation
between lagged dependent variable and the induced MA (1) error term. Also, a
necessary condition for a system GMM is that the error term is not serially correlated,
especially of second order, otherwise the standard errors of the instrument estimates
grow without bound. For this reason Arellano and Bond have developed a second
order autocorrelation test on which we base our analysis.
The SYS-GMM that was developed by Blundell and Bond (1998) added
additional necessary condition to the set of instruments proposed by Blundell and
152
Texas Tech University, Kolthoom Alkofahi, May 2014
Bover 1st-diff and SYS-GMM estimators. This condition states that, even if the
unobserved country-specific effect is correlated with the regressors‟ levels, it is not
correlated with their differences. The condition also means that the deviations of the
initial values of the independent variables from their long-run values are not
systematically related to the country-specific effects. For more details on these three
types of GMM estimators, please visit the work of CEL (1996), BHT (2001), and
Saima Nawaz (2011).
CROSS-COUNTRY GROWTH EXAMPLES USING GMM ESTIMATORS
The approach of 1st-diff GMM was introduced into the growth literature in the
contribution of CEL (1996). Since then similar techniques have been applied in
growth research by Levine et al. (2000), Forbes (2000), and Bond, Hoeffler and
Temple (BHT 2001) among others. CEL (1996), for instance, stated that there are two
sources of inconsistency in existing cross-country empirical work on growth, and
almost all the studies are plagued by at least one of these (the overwhelming majority
by both). First, the incorrect treatment of country-specific effects representing
differences in technology or tastes gives rise to omitted variable bias. In particular, it
is almost always assumed that such effects are uncorrelated with the other right-handside variables. They show that this assumption is necessarily violated due to the
dynamic nature of a growth regression. The second source is the endogeneity problem
that biases the estimate for economic growth regressions. Using the restricted and
unrestricted models of first difference dynamic panel GMM estimator for the Non-oil
sample set, CEL found very low estimates of capital share of income and a very fast
speed of convergence. They have found that these values in contrast with the
implications of the Solow growth model, and favors open economy versions of the
neoclassical growth model.
The work of CEL was revised by BHT (2001) who criticized the preceding
estimator for it may produce a serious drawback. They demonstrated that, the result of
153
Texas Tech University, Kolthoom Alkofahi, May 2014
the initial income, obtained by CEL, is likely to be seriously biased consistent with the
presence of weak instruments, known as large finite sample bias, and consequently,
biased the estimates of the rate of convergence and income per worker share with
respect to capital. BHT stressed out on the reasons of why the 1 st diff-GMM is
behaving poorly: (a) if the dependent variable follows a random walk, (b) if the
explanatory variables are persistent over time, and (c) if the time dimension of the
sample is small. Under these conditions, lagged levels of the variables are only week
instrument for the subsequent first difference.
BHT (2001) applied both types of GMM estimation of the Solow model and
augmented Solow model using same data set used by CEL, but only reported the
results of the unrestricted regression. To deal with the problems encountered when
using the 1st-diff, BHT proposed two possible solutions that rely on using more
informative set of instruments:
[1] The use of the system generalized methods of moments (SYS-GMM)
estimator suggested by Arellano and Bover (1995) and Blundell and Bond
(1998). This estimator exploits an assumption about the initial conditions
to obtain moment conditions that remain informative even for persistent
series. Based on their application, the additional instruments were highly
informative, and the results were found consistent with the Solow growth
framework; their results of the SYS-GMM indicated a rate of convergence
of around 2.1% a year, which is similar to the standard cross-section
finding.
[2] The use of external variable as another instrument to strengthen the
instrument set used for the equation of first differences. In their case, they
have used the lags of school enrollment as instruments in estimating the
basic growth model, since the augmented model version suggested that the
154
Texas Tech University, Kolthoom Alkofahi, May 2014
school enrolment can be omitted from their specification of the model.
Accordingly, the rate of convergence is found to equal 4.2%, nevertheless,
the Sargan test indicated that the lag of school enrolment may not be a
valid instruments.
At the end, unlike the conclusion of CEL that invalidated the Solow model for
the fast rate of convergence and low estimates of capital’s share, BHT results support
the Solow model’s implication and its usefulness of explaining income differences
across countries.
However, even though BHT recommended two solutions for subsequent growth
research, they give a possibility of the estimates to be imprecise, and biased due to
heterogeneity in the slope parameter that could invalidate the use of lagged values of
serially correlated regressors as instruments.
Another problem of using the SYS-GMM estimator can arise if the instruments
are too many, leading to over fitting of the model (Roodman, 2006). This turned out to
be the problem that we encountered when we first applied SYS-GMM based on
Blundell and Bond (1998), taking into account as many instruments as possible.
However, there is little guidance in the literature to determine how many instruments are
“too many” (Roodman 2006, Ruud 2000). A recommended rule of thumb by Roodman is
that instruments should not outnumber individuals (or countries). For this reason, we
apply the Blundell and Bond SYS-GMM conditional on fewer set of instruments to use
for the sake of being in conformity across all the samples included in this study.
155
Texas Tech University, Kolthoom Alkofahi, May 2014
ESTIMATING THE SOLOW GROWTH MODEL
To estimate the impact of FDI on economic growth, this study first employs a
dynamic panel data analysis model based on the 1st-DIFF GMM and SYS-GMM
estimators to test the validity of the Solow model. In attempt to realize if taking care
of the endogeneity and measurement error problems would produce more consistent
estimates of the parameters (α, λ), and later on , we estimate the Solow and the
augmented Solow model in restricted and unrestricted forms. We then compare the
results of estimated parameters to those obtained from cross-sectional and fixed effects
panel estimation. The reason why we are interested in such estimator is because;
advocates of the GMM estimator consider it immune to the inconsistency problems
that invalidate standard techniques. We then refer to CEL (1996) and BHT (2001)
methodologies and compare our findings for the Non-oil samples with their finding for
the same sample set. BHT (2001) did not report the restricted version of both the
Solow and the augmented Solow model, for that the estimated values of α is only
compared to those of CEL (1996). We hope to find plausible estimates of α and λ that
could put more weight on the importance of the role of the Solow model, and provide
more evidence on the effect that FDI exert on economic growth and the speed of
convergence.
To proceed with the analysis, we use same data structures as in the panel data
approach; that is we average the data for the control variables over non-overlapping,
five-year periods, so the data permits 6 observations for each country. This application
optimally utilizes and employs all the linear moment restrictions implied by a d-panel
model. As in CEL and BHT, all variables are expressed as deviations from time
means, which eliminates the need for time dummies.
156
Texas Tech University, Kolthoom Alkofahi, May 2014
DISCUSSION OF RESULTS BASED ON GMM ESTIMATION
Our results for the basic Solow growth model and its augmentation with FDI
are reported in Table IX.1 through TableIX.7 in restricted and unrestricted form,
where both of GMM estimators treat the control variables as potentially endogenous.
The first two columns report the results of estimating the Solow model using 1 st-Diff
and SYS-GMM estimators, and the other two columns report the results of the
augmented Solow model.
We start by analyzing the results obtained by 1 st-Diff GMM estimator for the
basic Solow model. The results show that there is a negative relationship between the
initial level of income per worker and subsequent growth rate. The estimated
coefficients of the saving rate and growth rate of working age population are negative
in sign but similar in magnitude for some samples. The restricted model is not rejected
at 95% significance level for most of the samples except the OECD and middle
income developing samples, however, the implied value of capital share is acceptable
and equals to 39% for both samples, accordingly, the rate of convergence is found to
be reasonable and equal to (3.60%, and 6.76%).
Nevertheless, the results obtained by 1 st-diff GMM estimator produce
unfavorable results. For example, this GMM estimator generates high rate of
convergence, and very low estimate of capital share with respect to capital for all the
samples except for the OECD and Intermediate samples. If we were to accept the
results obtained by this sort of GMM estimator, we would reject the basic Solow
model for reasons previously mentioned.
One way to deal with this problem is to discuss if adding FDI to the regression
equation would alter these results. For this reason, we implement the model with FDI
as another explanatory variable that is treated endogenous. We estimate FDI effect on
economic growth using both types of GMM estimator, the results of 1 st-diff GMM is
reported in the third column.
157
Texas Tech University, Kolthoom Alkofahi, May 2014
Unfortunately, the estimates produced by 1st-diff GMM estimator show that the effects
of FDI on economic growth is insignificant, not different from zero for all the samples
except the high income developing sample, and is negative for the middle income and
low income developing samples. This means that, 1st-diff GMM estimation is rejecting
the augmented Solow model as well, similar to the finding of CEL (1996).
In sum, the results that are reported in tables VIII.1 through VIII.7show that
the 1st –diff GMM estimator may be subject to a large downward finite sample bias,
especially since the number of time periods available is small. This suggests that,
even though we expect the inclusion of the current or lagged values of the regressors
in the instrument set would improve the behavior of the 1 st-diff GMM estimator, as in
some applications of the growth literature, the results of our application failed to meet
our expectation.
But how can we disclose if serious finite sample bias are existent?
BHT proposed that one indication can be obtained by comparing the 1 st- diff
GMM results to alternative estimates of the initial level of income per worker; for
example, the OLS and fixed effects panel estimations. The OLS is thought to give
estimates of the initial level of income that is biased upwards in the presence of
individual specific effects. On the other hand, the fixed effects estimation give an
estimate that might be seriously biased downward in short panel. Thus, the consistent
estimates of the initial level of income can be expected to lie in between the OLS level
and the fixed effects (within group) estimates. Likewise, a finding that the 1 st- diff
GMM estimates of the coefficient of the initial level of income lies close or below the
fixed effects estimates can be regarded as a signal that biases due to weak instruments
may be important. In this case, it is appropriate to employ SYS-GMM estimator since
it is considered as estimator that have superior finite sample properties, and is better
suited to estimating regressions with persistent panel data.
158
Texas Tech University, Kolthoom Alkofahi, May 2014
In order to detect if the finite sample bias are present in our analysis, we first
compare the initial values of income per worker generated by 1 st –diff GMM estimator
by those produced by cross-country frame work and fixed effects panel estimations
that are reported in Table IV ( A and B), Table V. (A and B), and Table VII (1 to 7).
For concreteness, we extract the coefficients of initial level of income per worker of
the restricted regression from the above mentioned tables them in Table IX. i, and
Table IX. ii. These tables would make it easier for us to compare the results instead of
going back in forth to find the coefficients we need. The reason why we only chose the
restricted regression is to make the argument meaningful, short, and if it is applicable
for the restricted regression, then it is applicable for the unrestricted regression as well.
For example, if we need to know if the 1 st-diff produces results that are subject
to finite sample bias in the Non-oil sample, we compare the coefficient produced by
1st-diff GMM of the restricted regression; i.e. ( -0.499) to values displayed in column
(1) Table IX. i. We can see that, the point estimate lies below the corresponding fixed
effects estimate (-0.313) which is likely to be seriously biased downwards in short
panel like this one. In fact, this is true for all the samples except the OECD sample,
where the 1st-diff GMM estimate is significant and equal to (-0.195) and the
corresponding fixed effects estimate is significant and equals to (-0.199).
One might see that, the OLS estimates for some samples lies below the fixed
effect estimates, however, for the sake of comparison, we only consider the values
obtained by the fixed effects estimates since it is more reliable for it correct the
unobserved country specific effects.
159
Texas Tech University, Kolthoom Alkofahi, May 2014
Table IX. i : list of
coefficients based on the basic Solow model (restricted regression)
Sample
Column #
Cross OLS
Non-oil
(1)
-0.073***
Intermediate
(2)
-0.111***
OECD
(3)
-0.445***
DEV all
(4)
-0.113*
High
(5)
-0.583***
Middle
(6)
-0.545***
Fixed
effects
-0.313***
-0.221***
-0.211***
-0.333***
-0.199***
-0.240***
Table IX. ii : list of
Low
(7)
-0.263
-0.419
coefficients based on Augmented Solow model (restricted regression)
Sample
Column #
Cross OLS
Non-oil
(1)
-0.069***
Intermediate
(2)
-0.114***
OECD
(3)
-0.457***
DEV all
(4)
-0.102
High
(5)
-0.663
Middle
(6)
-0.542***
Low
(7)
-0.227
Fixed
effects
-0.304***
-0.264***
-0.343***
-0.350***
-0.235***
-0.464***
-0.254***
160
Texas Tech University, Kolthoom Alkofahi, May 2014
The problem of finite sample bias is also present when we estimate the
augmented Solow model. The 1st-diff estimates lie below the corresponding fixed
effects estimates for all the samples except for the OECD sample, where the 1 st-diff
GMM estimate of the coefficient of initial level of income is significant and equals to
(-0.187) which lies above the corresponding fixed effects estimate of (-0.343).
In addition to that, when comparing our results of 1st –diff estimates to those obtained
by CEL (1996); we find that even though there is some similarities, our new
constructed, and extended data for the Non-oil sample dose not change the result for
good, our application of 1 st-diff GMM finnd that the implied values of initial level of
income per worker, λ , and α are (-0.499, 13.83%, 0.11), where the corresponding
estimates of CEL as shown in Table IX.iii are (-0.490, 11.96%, 0.104) respectively.
These values confirm that the sort of finite sample bias dose exist due to weak
instruments; therefore, our interest should be switched to the SYS-GMM estimation.
TableIX.iii: GMM estimations CEL and BHT, dependent variable is
CEL 1st -diff
Sample Non-oil
Solow Model
Basic
BHT SYS-GMM
Augm.
-0.473
------
Basic
Augm
-0.101
-0.081
(0.052)
(077)
(Unrestricted)
(0.079)
λ% a year
12.80
7.9
2.10
1.70
Sargan Test
0.31
------
0.43
0.26
p-value of Restriction
0.60
0.43
------
------
-0.490
-------
(Unrestricted)
(restricted)
(0.140)
λ % a year (Restricted)
13.50
6.79
------
------
Implied α
0.104
0.49
------
------
-0.259
------
------
------
------
------
Implied β
Sargan Test
0.15
161
Texas Tech University, Kolthoom Alkofahi, May 2014
We now test if the finite sample bias is present when the SYS-GMM estimator
is employed. The second and the fourth columns of Table IX.1 and those that follow
report the results from using SYS-GMM in estimating the basic and Augmented
Solow growth models. To know whether or not the problem of finite sample bias is
present, we check if the estimates of initial level of income produced by the SYSGMM estimator lie below the cross country OLS and above the fixed effects estimate.
To keep the study manageable, we only compare the results of the restricted
regression. For instance, the estimate of the initial level of Income per worker for the
Developing sample using SYS-GMM is significant and equals to (-0.184), this value
lies comfortably above the fixed effects estimate of (-0.333) and below the OLS cross
country estimate of (-0.113). The story is similar if we look at any sample other than
the OECD or the Low income developing sample. The basic Sargan test of overidentifying restrictions is not detecting any problem with instruments validity for any
of the samples.
We then test the validity of the Solow model based on the SYS-GMM
estimator. The results are displayed in column two of each table. We can see that the
results uncover the existence of a negative relationship between the initial level of
income per worker and subsequent growth rates for all the samples, in other word,
there is evidence that countries are converging to their respective steady states.
Another implication of the Solow model that is met regards the coefficients of the
saving rate and growth of working age population. These coefficients are opposite in
sign, and for some samples, equal in magnitude.
One finding that is also valid for all samples is that, the restricted model is not
rejected for all the samples confirming that countries at the end of the period are
converging to their respective steady states. The rate of conditional convergence for
the MRW samples are (2.11, 1.05, and 4.25) and for the developing and subdeveloping samples λ equals to (4.06, 3.23, 6.36, and 9.45). Accordingly, λ for the
162
Texas Tech University, Kolthoom Alkofahi, May 2014
sub- developing samples is found the highest for the low income developing samples
and the lowest for the high income developing samples, which implies that countries
are converging in terms of the two types of convergence; absolute convergence, and
conditional convergence.
One striking result we can extract is that, the Non-oil sample is converging to
its steady state at a rate equals to 2.11%. The coefficients of saving rate and working
age population in BHT were found as (0.188, -0.309), our corresponding estimates for
the same sample are found as (0.178, -0.174), our estimated coefficients are opposite
in sign, and without any doubt, equal in magnitude. BHT accept the validity of the
Solow model regardless the value of capital share of Income. If we were to follow
their steps, we would accept the Solow model for most of the samples. Yet, in spite of
their results, we insist on analyzing the restricted regression as well. We find that, the
implied value of capital share of income (α) differs from one third for most of the
samples. For the MRW sample, α equals to (0.65, 0.72, 0.38) and the corresponding
estimates for the developing and sub-developing samples is (0.50, 0.62, 0.32, 0.14).
We can see that the value of capital share meets the standard value of one third only
for the OECD and the middle income developing samples. Frankly speaking, we can’t
reject the Solow model for these two samples, which answers one of our questions
that, grouping countries does matter; the validity of the Solow model is affected by the
way that samples are constructed.
The extension that our study is built on is to incorporate the FDI into the
model. The fourth column of each table represents the results of analyzing the
augmented Solow model using SYS-GMM estimator. We can see that FDI positively
enters the growth equation for all the samples, though it is only significant for the high
income developing sample, where the inclusion of FDI helps accelerate the
convergence rate from (3.23) to (3.43), and lower the estimate of α from 0.62 to 0.44.
The implied value of income share with respect to FDI activities is found0.10; this is
the highest implied value amongst the set of samples. FDI also help accelerate
163
Texas Tech University, Kolthoom Alkofahi, May 2014
economic convergence for the Intermediate sample and a slight increase occurred to
the OECD samples; the convergence rate is now (1.65% , 4.25%) a year compared to
the rate found by the basic Solow model of (1.05%, 4.29) a year. For the rest of the
samples, FDI decreased the rate of convergence; we can reconcile this finding by
referring to the FDI as development engine, where it helps those economies to perform
well specially of the samples that are converging at a faster rate of convergence. We
might also qualify this finding and acknowledge that that there is a great deal of
uncertainty in measuring convergence rates as was emphasized by Nerlove (2000).
Our last comment in this regard is that, the share of income with respect to FDI
activities ( ) ranges from 0.02 for the Low income sample to 0.10 for the high income
developing sample. This finding in match to the results obtained by the fixed effects
estimation produced in Table VII.1 through Table VII.7. And finally, the Sargan tests
of over identification dose not detect any problem for any of the samples. This means
that the extra instruments that are used are informative, do make a substantial
difference to the 1st-diff GMM results, and it increases the precision of the SYS-GMM
estimates. One more thing, based on the results of restricted regression, we conclude
that we fail to reject the augmented Solow model for the high income developing and
the OECD samples.
164
Texas Tech University, Kolthoom Alkofahi, May 2014
COMMENTS ON USING GMM ESTIMATOR
1. Overall, the results obtained in our analysis suggest that the 1st-Diff GMM
estimates are subject to serious finite sample bias due to weak instruments.
2. The problem of finite sample bias can be addressed using the SYS-GMM
estimator that yields a considerable improvement in precision compared to 1 st-diff
GMM estimator.
3. Finding evidence of convergence is thought as a tool to support the validity of the
Solow model, however, it is hard to reject the Solow model or the augmented
Solow model just for the high estimates of capital share of income. The validity of
the Solow model greatly depends on the way samples are constructed; data for the
samples that include more homogenous countries tend deliver more consistent
results.
4. The inclusion of FDI positively affects the growth rate of income per worker for
all the samples. Even though FDI is found to accelerate the rate of convergence for
some samples, other samples’ rate of convergence slows down by the inclusion of
FDI, which means that instead of allowing lower income countries to catch up to
higher income countries, FDI is widening the gap between countries, because of
differential effects: If high income countries can maintain or increase their net
inflows of FDI as a percentage of GDP, they can experience a higher growth rate
as they can utilize the capital efficiently, hence, increasing the income gap.
5. In spite of that, the income share with respect to FDI activities ranges from 0.02
for low income developing countries to 0.10 for high income developing countries.
6. Even though our results support the SYS-GMM estimator over the 1st-diff
estimator, we think that there exist some sort of problem that prevents the results
165
Texas Tech University, Kolthoom Alkofahi, May 2014
from being robust, as BHT give a possibility of the estimates to be imprecise, and
biased due to heterogeneity in the slope parameter that could invalidate the use of
lagged values of serially correlated regressors as instruments.
166
Texas Tech University, Kolthoom Alkofahi, May 2014
TABLE IX.1 GMM Estimation: Non-Oil Sample
ESTIMATION
SOLOW MODEL
AUGMENTED SOLOW MODEL
GMM TYPE
1ST –DIFF
SYS
1ST –DIFF
SYS
# OF OBSERVATION
420
504
410
494
-0.483
(0.349)
0.075
(0.050)
-0.038
(0.073)
0.003
(0.008)
13.21
0.00
18
-0.101***
(0.046)
0.185
(0.118)
-0.154
(0.128)
0.008
(0.009)
2.14
0.06
23
-0.437*
(0.302)
0.051
(0.051)
0.003
(0.007)
0.64
11.51
0.10
0.01
0.01
17
-0.098***
(0.076)
0.180**
(0.081)
0.007
(0.008)
0.84
2.06
0.63
0.03
0.06
22
**
-0.537
(0.217)
0.089**
(0.043)
-0.026
(0.055)
λ% in a year
Sargan Test
# of instrument
p-value
λ% in a year
α
β
Sargan Test
# of instrument
15.39
0.00
17
UNRESTRICTED
-0.115***
(0.041)
0.178
(0.132)
-0.174
(0.122)
-0.499**
(0.228)
0.062
(0.056)
2.3
0.09
22
RESTRICTED
-0.105***
(0.027)
0.185**
(0.084)
0.40
13.83
0.11
0.98
2.11
0.65
0.01
16
0.07
21
NOTE: dependent variable is
167
Texas Tech University, Kolthoom Alkofahi, May 2014
TABLE IX.2 GMM Estimation: Intermediate Sample
ESTIMATION
GMM TYPE
# OF OBSERVATION
λ% in a year
Sargan Test
# of instrument
p-value
λ% in a year
α
β
Sargan Test
# of instrument
SOLOW MODEL
1 –DIFF
SYS
370
444
UNRESTRICTED
***
-0.501
-0.066**
(0.133)
(0.033)
**
0.100
0.144***
(0.040)
(0.049)
-0.050
-0.135
(0.070)
(0.104)
ST
11.59
0.003
17
-0.512***
(0.160)
0.075
(0.043)
1.13
0.08
22
RESTRICTED
-0.061**
(0.030)
0.158***
(0.044)
0.60
11.96
0.13
0.94
1.05
0.72
0.003
16
0.06
21
NOTE: dependent variable is
168
AUGMENTED SOLOW MODEL
1ST –DIFF
SYS
362
436
-0.487***
(0.141)
0.090*
(0.047)
-0.047
(0.076)
0.001
(0.009)
11.12
0.02
18
-0.059***
(0.030)
0.149***
(0.054)
-0.138
(0.119)
0.010
(0.007)
1.01
0.02
23
-0.487***
(0.171)
0.068
(0.044)
-0.001
(0.009)
0.68
13.96
0.12
-------0.02
17
-0.051***
(0.022)
0.157***
(0.048)
0.009
(0.007)
0.88
1.65
0.73
0.04
0.08
22
Texas Tech University, Kolthoom Alkofahi, May 2014
TABLE IX.3 GMM Estimation: OECD Sample
ESTIMATION
SOLOW MODEL
AUGMENTED SOLOW MODEL
GMM TYPE
1ST –DIFF
SYS
1ST –DIFF
SYS
# OF OBSERVATION
120
144
119
143
-0.136***
(0.029)
-0.052
(0.080)
-0.169***
(0.061)
-0.005
(0.009)
2.44
0.61
18
-0.227***
(0.011)
0.077
(0.059)
-0.206**
(0.085)
0.002
(0.011)
4.29
0.28
23
-0.187**
(0.092)
0.129**
(0.051)
0.001
(0.009)
0.02
4.15
0.40
0.003
0.58
17
-0.227***
(0.007)
0.149**
(0.063)
0.006
(0.010)
0.08
4.29
0.39
0.02
0.29
22
***
-0.144
(0.019)
-0.410
(0.077)
-0.162***
(0.056)
λ% in a year
Sargan Test
# of instrument
p-value
λ% in a year
α
β
Sargan Test
# of instrument
2.59
0.73
17
UNRESTRICTED
-0.227***
(0.097)
0.070
(0.052)
-0.196**
(0.081)
-0.195**
(0.092)
0.125**
(0.058)
4.29
0.40
22
RESTRICTED
-0.225**
(0.102)
0.136**
(0.068)
0.02
3.6
0.39
0.06
4.25
0.38
0.65
16
0.40
21
NOTE: dependent variable is
169
Texas Tech University, Kolthoom Alkofahi, May 2014
TABLE IX.4 GMM Estimation: Developing Income Sample
ESTIMATION
SOLOW MODEL
AUGMENTED SOLOW MODEL
GMM TYPE
1ST –DIFF
SYS
1ST –DIFF
SYS
# OF OBSERVATION
320
384
306
370
-0.660**
(0.369)
0.110***
(0.071)
-0.04
(0.040)
0.010
(0.014)
2.16
0.02
18
-0.194***
(0.025)
0.208**
(0.125)
0.043
(0.072)
0.022
(0.015)
4.31
0.06
23
-0.672**
(0.450)
0.058
(0.049)
0.010
(0.014)
0.08
2.43
0.08
0.01
0.01
17
-0.170***
(0.089)
0.170**
(0.066)
0.023*
(0.013)
0.09
3.74
0.47
0.06
0.14
22
-0.821
(0.844)
0.093*
(0.050)
-0.024
(0.055)
λ% in a year
Sargan Test
# of instrument
p-value
λ% in a year
α
β
Sargan Test
# of instrument
34.4
0.02
17
UNRESTRICTED
-0.216***
(0.051)
0.233**
(0.110)
-0.061
(0.043)
-0.783
(0.625)
0.053
(0.039)
4.86
0.09
22
RESTRICTED
-0.184***
(0.021)
0.184***
(0.065)
0.17
30.55
0.06
0.15
4.06
0.50
0.01
16
0.16
21
NOTE: dependent variable is
170
Texas Tech University, Kolthoom Alkofahi, May 2014
TABLE IX.5 GMM Estimation: High Income Developing Sample
ESTIMATION
SOLOW MODEL
AUGMENTED SOLOW MODEL
GMM TYPE
1ST –DIFF
SYS
1ST –DIFF
SYS
# OF OBSERVATION
70
84
68
82
-0.421**
(0.200)
0.128**
(0.061)
-0.234*
(0.141)
0.035**
(0.016)
9.10
0.79
18
-0.123***
(0.029)
0.222***
(0.072)
-0.307***
(0.098)
0.035**
(0.007)
2.19
0.92
23
-0.524***
(0.168)
0.147**
(0.066)
0.033*
(0.018)
0.58
14.98
0.21
0.05
0.59
17
-0.157***
(0.021)
0.227***
(0.075)
0.035*
(0.018)
0.88
3.43
0.44
0.10
0.14
22
***
-0.514
(0.176)
0.121*
(0.067)
-0.260*
(0.136)
λ% in a year
Sargan Test
# of instrument
p-value
λ% in a year
α
β
Sargan Test
# of instrument
14.43
0.57
17
UNRESTRICTED
-0.176***
(0.020)
0.248**
(0.106)
-0.274**
(0.104)
-0.530***
(0.172)
0.175*
(0.102)
3.87
0.82
22
RESTRICTED
-0.149***
(0.030)
0.246***
(0.080)
0.24
15.10
0.25
0.83
3.23
0.62
0.51
16
0.83
21
NOTE: dependent variable is
171
Texas Tech University, Kolthoom Alkofahi, May 2014
TABLE IX.6 GMM Estimation: Middle Income Developing Sample
ESTIMATION
SOLOW MODEL
AUGMENTED SOLOW MODEL
GMM TYPE
1ST –DIFF
SYS
1ST –DIFF
SYS
# OF OBSERVATION
100
120
99
119
-0.309***
(0.118)
0.123**
(0.056)
-0.282***
(0.046)
-0.002
(0.015)
7.39
0.36
18
-0.228***
(0.069)
0.078***
(0.037)
-0.278***
(0.072)
0.015
(0.017)
2.19
0.51
23
-0.280***
(0.106)
0.185***
(0.052)
-0.001
(0.015)
0.02
6.58
0.40
0.00
0.30
17
-0.237***
(0.068)
0.114***
(0.028)
0.019
(0.014)
0.04
5.42
0.31
0.05
0.56
22
***
-0.311
(0.105)
0.123**
(0.055)
-0.283*
(0.047)
λ% in a year
Sargan Test
# of instrument
p-value
λ% in a year
α
β
Sargan Test
# of instrument
7.46
0.37
17
UNRESTRICTED
-0.254***
(0.070)
0.085***
(0.032)
-0.300**
(0.077)
-0.287***
(0.111)
0.185***
(0.046)
5.85
0.48
22
RESTRICTED
-0.273***
(0.065)
0.128***
(0.029)
0.03
6.76
0.39
0.05
6.36
0.32
0.30
16
0.51
21
NOTE: dependent variable is
172
Texas Tech University, Kolthoom Alkofahi, May 2014
TABLE IX.7 GMM Estimation: Low Income Developing Sample
ESTIMATION
SOLOW MODEL
AUGMENTED SOLOW MODEL
GMM TYPE
1ST –DIFF
SYS
1ST –DIFF
SYS
# OF OBSERVATION
150
180
139
169
-0.678
(0.457)
0.064
(0.071)
-0.015
(0.032)
0.024
(0.016)
2.27
0.57
18
-0.300***
(0.097)
0.114*
(0.066)
0.058
(0.087)
0.018
(0.013)
7.10
0.34
23
-0.704***
(0.975)
0.006
(0.030)
0.022
(0.014)
0.18
2.43
0.01
0.03
0.57
17
-0.282***
(0.013)
0.059
(0.047)
0.017
(0.016)
0.10
6.62
0.06
0.02
0.28
22
-1.068
(0.844)
0.076*
(0.045)
-0.003*
(0.064)
λ% in a year
Sargan Test
# of instrument
p-value
λ% in a year
α
β
Sargan Test
# of instrument
-----0.50
17
UNRESTRICTED
-0.365***
(0.199)
0.124*
(0.072)
0.082
(0.121)
-1.051
(0.908)
0.019
(0.029)
9.09
0.41
22
RESTRICTED
-0.376***
(0.139)
0.059
(0.062)
0.31
-0.02
0.16
9.45
0.14
0.23
16
0.22
21
NOTE: dependent variable is
173
Texas Tech University, Kolthoom Alkofahi, May 2014
CHAPTER X
FINAL CONCLUSIONS
This paper investigates the impact of foreign direct investment on the level of
income per capita and its economic growth using three sets of techniques; crosscountry OLS framework, panel data approach (both pooled OLS and fixed effects),
and dynamic panel data estimator in the form of first-differenced and system
generalized methods of moments. Previous studies have either assessed the validity of
the Solow model or the impact of foreign direct investment separately. In this paper,
we assess the validity of the Solow model in the presence of foreign direct investment.
The study starts by analyzing the work of Mankiw, Romer, and Weil (1992) using
their samples with extended, new, and revised data. It later adds the developing
countries and sub-sections it into three sub-samples. The study investigates the
validity of the Solow model and its augmentation with foreign direct investment over
the period 1980-2010. It then employs a panel data technique to further support the
Solow model’s implication, especially that panel data takes care of the country
specific effects that is ignored by the OLS frame-work. Considering out of steady state
behavior, this study takes the work of Islam (1995) seriously, and study the
conditional convergence of the basic and augmented Solow models. The results of the
panel estimation produce more efficient and reliable results compared to the crosscountry OLS framework.
This paper contributes to the literature by finding evidence of conditional
convergence that is considered a way to support the Solow growth model’s validity
and refutes the endogenous version of the growth theory. It also contributes to the
existing literature by finding evidence of the great positive influence that FDI exert on
the level of income and its economic growth; based on panel data estimation, FDI is
proved to act as a growth enhancement engine especially for countries with higher
income per worker or higher stage of development.
174
Texas Tech University, Kolthoom Alkofahi, May 2014
This study not only finds evidence of conditional convergence, it also finds
that countries with lower income per worker experience a higher increase in economic
growth towards their respective steady states compared to countries with higher level
of income per worker. The speed of convergence is influenced by the net inflows of
foreign direct investment, which allow poorer countries to catch up with the richer
countries at a faster rate of convergence.
We further employ dynamic panel data study that is considered, by many, as
the most efficient approach since it takes care of endogeneity problem, and perhaps,
measurement error. These two issues were not addressed by fixed effect estimates.
Hence, our study confirms the finding of CEL (1996) whose work employs the 1 stdifferenced version of GMM estimation, and system GMM estimator that was
employed by BHT (2001). Overall, the results obtained in our analysis suggest that the
1st-Diff GMM estimates are subject to serious finite sample bias due to weak
instruments, and this issue is addressed by employing the SYS-GMM estimations.
Even though our favorable results are those obtained by SYS-GMM, this estimator
underestimates the role that FDI plays in some economies. It also focuses on
enhancing the development of high income developing ample. One way to uncover the
reasons of such results is to follow the argument of BHT where they gave the
possibility of the estimates to be imprecise, and biased due to heterogeneity in the
slope parameter that could invalidate the use of lagged values of serially correlated
regressors as instruments.
Major policy implications can be made based in the results of this study. Since
FDI enhances the growth rates of income per worker, the governments of those
countries shall keep providing incentives and lowers the barriers to encourage other
countries to invest domestically. Favorable political and macroeconomic conditions,
better environments, political stabilities, legislations concerning the stability and the
protection of foreign investments, and tax incentives shall be enforced.
175
Texas Tech University, Kolthoom Alkofahi, May 2014
Finally, the results of this paper suggest some direction for further research. In
this study, we only incorporate FDI as another factor of inputs; the next step is not
only include the human capital accumulation as another factor of input, also assess if
the interaction of FDI and human capital accumulation would have anything to add to
the growth of income per worker. In this paper, we focus on the closed economy
version of the Solow model; future research should focus on studying the open
economy version of the Solow growth model, and determine what would be a better
way to study the effect of FDI on economic growth.
176
Texas Tech University, Kolthoom Alkofahi, May 2014
BIBLIOGRAPHY
Adams, Samuel. "Can foreign firect investment (FDI) Help to promot growth in
Africa?" African Journal of Business Management Vol.3(5), (2009): 178-183.
Agrawal, Pradeep. ""Economic Impact Of Foreign Direct Investment In South Asia"."
Indira Gandhi Institute of Development Research (January 2000).
Alan Heston, Robert Summers and Bettina Aten. Penn World Table Version 7.1,
Center for International Comparisons of Production, Income and Prices at the
University of Pennsylvania. July 2012.
Alfaro, Laura. " “Foreign Direct Investment and Growth: Does the Sector Matter?”."
Harvard Business School, Boston (April 2003).
—. ""Foreign Direct Investment and Growth"." Harvard business School (April 2003).
Aviral Kumar Tiwari, and Mihai Mutascu. ""Economic Growth and FDI in Asia:A
Panel-Data Approach"." Economic Analysis & Policy, Vol. 41 No. 2,
(september 2011): pp. 173-187.
Ayanwale, Adeolu B. "" FDI and Economic Growth:Evidence from Nigeria"." AERC
Research Paper 165. Nairobi: African Economic Research
Consortium,Nairobi, April 2007.
Barro, robert J. "" Determinants of Economic Growth" A Cross-Country Empirical
Study"." NBER Working paper 5698 (Augest 1996).
Bond, S., Hoeffler, A. and Temple, J. "“GMM estimation of empirical growth
models”." Discussion Paper 3048, Centre for Economic Policy Research
(2001).
Bond, Stephen R., Hoeffler, Anke. And Temple, Jonathan. "“GMM Estimation of
Empirical Growth Models”." CEPR discussion paper series N 3048 (2001).
177
Texas Tech University, Kolthoom Alkofahi, May 2014
Borensztein, E., J. De gregorio and J.W Lee. ""How does FDI affect economic
growth"." Journal of International Economics (1998): 45(1):115-135.
Caselli, F., G. Esquivel and F. Lefort. ""Reopening the convergence debate: A new
look at cross-country growth empirics"." Journal of Economic Growth (1996):
1(3): 363-389.
De Mello, Luiz. "“Foreign Direct Investment-Led Growth: Evidence from Time Series
and panel data"." Oxford Economic papers (1999): 51:133-151.
Dewan, E and S. Hussein. "" Determinants of Economic Growth"." Working Paper,
Reserve Bank of Fiji (2001).
Di Liberto, A. "" Convergence and divergence in neoclassical growth models with
human capital"." Centre for North South Economic Research, University of
Cagliari and Sassari, Sardinia. (2005): CRENoS Working Paper no. 8.
Doythc, N., and Uctum, M. ""Does the worldwide shift of FDI from manufacturing to
services accelerate economic growth? " A GMM estimation study"." Journal
of International Money and Finance (2011): 30:410–427.
Freckleton, M., Wright, A. S. and Craigwell, R. " “Foreign Direct Investment,
Economic Growth and Corruption in Developing Economies”." University of
the West Indies, mimeo, August (2010).
Hak, Donna Chan Wah. "Do Foreign Direct Investment and Trade Openness
Accelerate Economic Growth?" The Honor Program (April,2011).
Islam, Nasrul. ""Growth empirics: a panel data approach"." The Quarterly Journal of
Economics (1995): 110, 4, November, pp.1127-71.
Kalbasi, Nahid. ""Economic growth convergence among Middle East Countries"."
Journal of Economics and International Finance (October 2010): Vol. 2(10),
pp. 231-236.
178
Texas Tech University, Kolthoom Alkofahi, May 2014
Kim, K., & Bang, H. "" The Impact of Foreign Direct Investment on Economic
Growth: A case study of Ireland "." Korea Institute of International Economic
Policy (2008): Working Paper 08-04 .
MADDALA, G.S. "" On the Use of Panel Data Methods with Cross-Country Data"."
Annales D’économie Et De Statistique (1999): N° 55-56.
Mankiw, Romer and Weil. "" A Contribution to the empirics of of economic growth."
Quarterly Journal of Economics (1992): 107(2):407-437.
Mathur, Somesh K. "Economic Growth and Conditional Convergence: Its speed for
Selected Regions for 1961-2001." Jamia Millia Islamia, New Delhi, India
(2005).
Michie, Jonathan. "“The Impact of Foreign Direct Investment on Human Capital
Enhancement in Developing Countries.”." A Report of the OECD, 1st
Draft,12th November (2001).
Moosa, Imad A. " Foreign Direct Investment Theory, Evidence and Practice". 9-62002.
Mora,Rui and Forte, Rosa. "“The Effects of Foreign Direct Investment on the Host
Country Economic Growth-Theory and Empirical Evidence"." (2009).
Mustafa Seref Akin and Valerica Vlad. "" The Relationship between Education and
Foreign Direct Investment: Testing the Inverse U Shape"." European Journal
of Economic and Political Studies (2011): No(4), pp 27-46.
OECD. " OECD Benchmark Definition of Foreign Direct Investment" fourth edition.
2008.
Ozturk, Ilhan and Kalyoncu, Huseyin. ""Foreign Direct Investment and Growth:An
Empiricial Investigation Based on Cross-Country Comparison"." ECONOMIA
INTERNAZIONALE (February 2007): Volume 60, No. 1, 75-82.
179
Texas Tech University, Kolthoom Alkofahi, May 2014
Romer, Paul. ""Idea gaps and object gaps in economic development"." Journal of
Monetary Economics (1993): 32: 543-573.
Silvio Contessi and Ariel Weinberger. ""Foreign Direct Investment, Productivity,and
Country Growth: An Overview"." Federal Reserve Bank of St. Louis Review
(March/April 2009): 91(2), pp. 61-78.
Syed T. Jawaid, Syed A. Raza. "Foreign Direct Investment, Growth and Convergence
Hypothesis: A cross Country Analysis." Munich Personal RePec Archive (May
2012).
UNCTAD. "" World Investment Report: FDI policies For Development:National and
International Perspectives"." United Nations, New York and Geneva (2003).
—. "" World Investment Report: Towards a New Generation of Investment
Polocies"." United Nations. New York and Geneva, 2012.
—. "“World Investment Report: Transnational Corporations and the Infrastructure
Challenge,” New York." United Nations (2009).
UNCTAD, 2011. ""World Investment Report: Non-Equity Modes Of International
Production And Development"." (n.d.).
Usha Nair-Reichert and Diana Weinhold. ""Causality Tests for Cross-Country Panels:
New Look at FDI and Economic Growth in Developing Countries"." Oxford
Bulletin of Economics and Statistics (2001): 63(2):153-171.
Whelan, Karl. "" The Solow Model of Economic Growth"." EC4010 Notes. 2005.
World Bank. "Statistics Database ." (2010).
YIGIT, MERVE. "" Employment Effects Of Foreign Direct Investment In The
Turkish Manufacturing Industry"." Department of Economics, Bilkent
University, Ankara (September 2010).
180
Texas Tech University, Kolthoom Alkofahi, May 2014
Zhuang, Hong. "" Foreign Direct Investment and Human Capital Accumulation in
China"." International Research Journal of Finance and Economics - Issue 19
(2008) (2008).
181
Texas Tech University, Kolthoom Alkofahi, May 2014
APPENDIX
Table X.1: COUNTRIES IN THE STUDY SAMPLES OF MRW
#
Non-oil
Intermediate income
OECD
1
Argentina
1
0
2
Australia
1
1
3
Austria
1
1
4
Bangladesh
1
0
5
Benin
1
0
6
Bolivia
1
0
7
Brazil
1
0
8
Burkina Faso
1
0
9
Cameroon
1
0
10
Canada
1
1
11
Central African Republic
0
0
12
Chad
0
0
13
Chile
1
1
14
Colombia
1
0
15
Congo, Dem. Rep.
0
0
16
Congo, Republic of
1
0
17
Costa Rica
1
0
18
Cote d`Ivoire
1
0
19
Denmark
1
1
20
Dominican Republic
1
0
21
Ecuador
1
0
22
Egypt
1
0
23
El Salvador
1
0
24
Finland
1
1
25
France
1
1
26
Germany
1
1
27
Ghana
1
0
28
Greece
1
1
29
Guatemala
1
0
30
Honduras
1
0
182
Texas Tech University, Kolthoom Alkofahi, May 2014
Table X.1. Continued
#
Non-oil
Intermediate income
OECD
31
India
1
0
32
Indonesia
1
0
33
Ireland
1
1
34
Israel
1
1
35
1
1
36
Italy
Jamaica
1
0
37
Japan
1
1
38
Jordan
1
0
39
Kenya
1
0
40
Korea, Republic of
1
1
41
Lesotho
0
0
42
Liberia
0
0
43
Madagascar
1
0
44
Malawi
1
0
45
Malaysia
1
0
46
Mali
1
0
47
Mauritania
1
0
48
Mexico
1
1
49
Morocco
1
0
50
Mozambique
1
0
51
Netherlands
1
1
52
New Zealand
1
1
53
Nicaragua
1
0
54
Niger
0
0
55
Nigeria
1
0
56
Norway
1
1
57
Pakistan
1
0
58
Panama
1
0
59
Papua New Guinea
1
0
60
Paraguay
1
0
183
Texas Tech University, Kolthoom Alkofahi, May 2014
Table X.1. Continued
#
Non-oil
Intermediate income
OECD
61
Peru
1
0
62
Philippines
1
0
63
Portugal
1
1
64
Rwanda
1
0
65
Senegal
1
0
66
Sierra Leone
1
0
67
Singapore
1
0
68
South Africa
1
0
69
Spain
1
1
70
Sri Lanka
1
0
71
Sudan
0
0
72
Sweden
1
1
73
Syria
1
0
74
Thailand
1
0
75
Togo
0
0
76
Trinidad &Tobago
1
0
77
Tunisia
1
0
78
Turkey
1
1
79
United Kingdom
1
1
80
United States
1
1
81
Uruguay
1
0
82
Venezuela
1
0
83
Zambia
1
0
84
Zimbabwe
1
0
184
Texas Tech University, Kolthoom Alkofahi, May 2014
Table X.2: DEVELOPING COUNTRIES AND SUB-SAMPLES
#
All Developing countries
High income
Middle Income
Low Income
1
Argentina
1
0
0
2
Bangladesh
0
0
1
3
Benin
0
0
1
4
Bolivia
0
1
0
5
Brazil
1
0
0
6
Burkina Faso
0
0
1
7
Cameroon
0
0
1
8
Central African Republic
0
0
1
9
Chad
0
0
1
10
Chile
1
0
0
11
Colombia
0
1
0
12
Dem. Rep. of the Congo
0
0
1
13
Congo
0
1
0
14
Costa Rica
1
0
0
15
Côte d'Ivoire
0
0
1
16
Dominican Republic
0
1
0
17
Ecuador
0
1
0
18
Egypt
0
1
0
19
El Salvador
0
1
0
20
Ghana
0
0
1
21
Guatemala
0
1
0
22
Honduras
0
1
0
23
India
0
0
1
24
Indonesia
0
1
0
25
Jamaica
0
1
0
26
Jordan
0
1
0
27
Kenya
0
0
1
28
Korea, Republic of
1
0
0
29
Lesotho
0
0
1
30
Liberia
0
0
1
31
Madagascar
0
0
1
32
Malawi
0
0
1
185
Texas Tech University, Kolthoom Alkofahi, May 2014
Table X.2. Continued
#
All Developing countries
High income
Middle Income
Low Income
33
Malaysia
1
0
0
34
Mali
0
0
1
35
Mauritania
0
0
1
36
Mexico
1
0
0
37
Morocco
0
1
0
38
Mozambique
0
0
1
39
Nicaragua
0
0
1
40
Niger
0
0
1
41
Nigeria
0
0
1
42
Pakistan
0
0
1
43
Panama
1
0
0
44
Papua New Guinea
0
0
1
45
Paraguay
0
1
0
46
Peru
0
1
0
47
Philippines
0
1
0
48
Rwanda
0
0
1
49
Senegal
0
0
1
50
Sierra Leone
0
0
1
51
Singapore
1
0
0
52
South Africa
1
0
0
53
Sri Lanka
0
1
0
54
Sudan
0
0
1
55
Syrian Arab Republic
0
1
0
56
Thailand
0
1
0
57
Togo
0
0
1
58
Trinidad and Tobago
1
0
0
59
Tunisia
0
1
0
60
Turkey
1
0
0
61
Uruguay
1
0
0
62
Venezuela
1
0
0
63
Zambia
0
0
1
186

Econ. 771: Problem Set I

bio-technology - Vignan University

Advanced Macroeconomics

Determinants of Foreign Direct Investment: Empirical evidence from

View agenda

OLS Brochure

Result of Placement Drive by Horizon Telecom

84(R) HR 183 - Introduced version

la porte resident claims $1 million scratch- off top prize

UNIVERSITY OF KERALA

Young better equipped to teach, feel experts at IIT