Privatization and Economic Growth

This article was downloaded by:[Hiroshima University]
On: 25 July 2008

Access Details: [subscription number 789277225]
Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954
Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Journal of Development Studies

Publication details, including instructions for authors and subscription information:
http://www.informaworld.com/smpp/title~content=t713395137
Privatisation and economic growth in developing
countries
Paul Cook a; Yuichiro Uchida a
a
Centre on Regulation and Competition, Institute for Development Policy and
Management, University of Manchester, UK.
Online Publication Date: 01 August 2003

To cite this Article: Cook, Paul and Uchida, Yuichiro (2003) 'Privatisation and
economic growth in developing countries', Journal of Development Studies, 39:6,
121 — 154
To link to this article: DOI: 10.1080/00220380312331293607
URL: http://dx.doi.org/10.1080/00220380312331293607
PLEASE SCROLL DOWN FOR ARTICLE
Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf

This article maybe used for research, teaching and private study purposes. Any substantial or systematic reproduction,
re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expressly
forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents will be
complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should be
independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings,
demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or
arising out of the use of this material.
Downloaded By: [Hiroshima University] At: 07:16 25 July 2008 396jds05.qxd 29/08/03 14:13 Page 121
Privatisation and Economic Growth in

Developing Countries
PA U L C O O K and Y U I C H I R O U C H I D A
This article re-examines the relation between privatisation and

economic growth. Previous studies that have attempted to measure
this relationship have concluded that privatisation has had a
sizeable positive effect on economic growth. Our study uses data
for 63 developing countries over the time period 1988–97. It uses
the framework of an extreme-bounds analysis (EBA) to conduct a
cross-country growth regression analysis. Our findings contradict
earlier results, but reaffirm the view that effective competition and
its regulation may need to accompany privatisation to make a
positive impact on economic growth.
I. INTRODUCTION
Privatisation accelerated in the 1990s. In Europe this was sparked by the

liberalisation of markets at the European Union level and budgetary
constraints faced by government [Parker, 1999]. Privatisation particularly
gained momentum in the late 1980s and spread to a wide range of
developing economies. Over the last decade a significant proportion of
privatisation transactions have been in developing economies and have
entailed sales of public utilities. Privatisation transactions for the utilities
sector have accounted for over a third of all transactions in developing
economies since 1988. In the period 1988–93 the value of sales for
infrastructure industries amounted to US$30 billion, compared to US$78
billion for all privatisation transactions in developing economies [World
Bank, 1995a, Kikeri, 1998].
There is now sufficient time lapse since privatisation was first
implemented in developing countries for the results to be evaluated at the
macroeconomic level. Admittedly, privatisation proceeded in the 1980s
Paul Cook and Yuichiro Uchida, Centre on Regulation and Competition, Institute for
Development Policy and Management, University of Manchester, UK; correspondence: yuichiro-
uchida@man.ac.uk.
The Journal of Development Studies, Vol.39, No.6, August 2003, pp.121–154
PUBLISHED BY FRANK CASS, LONDON
396jds05.qxd 29/08/03 14:13 Page 122
122 THE J OURNA L O F D E V E L O P ME N T S T U D I E S

Downloaded By: [Hiroshima University] At: 07:16 25 July 2008
without much knowledge of either its impact or contribution to economic

growth. Instead greater attention was paid to measuring the extent of
privatisation [Bennell, 1997]. This situation is changing and more recent
contributions assessing post-privatisation performance are reviewed in
Megginson and Netter [2001]. Among those most frequently cited are:
Galal, Jones, Tandon and Vogelsang [1994]; Megginson, Nash and Van
Randenborgh [1994] and Boubakri and Cosset [1998]. However, these
studies have largely concentrated on financial measures of performance,
and have incorporated industrialised countries and higher income
developing countries in their relatively small databases. In recent years
there have also been a few attempts to measure the direct impact of
privatisation on economic growth in a cross-country context [Plane, 1997;
Barnett, 2000]. These studies have concluded that privatisation has had a
sizeable positive effect on economic growth.
The purpose of this analysis is to re-examine the relation between
privatisation and economic growth, using data for 63 developing countries
over the period 1988–97. The next section provides a brief review of the
theoretical links between a change in ownership from public to private and
economic growth, and of the empirical literature that has examined this
relationship. The third section discusses the methods adopted to explore the
relation between privatisation and economic growth and the data used. The
fourth section discusses the results of the analysis. The final section
provides a summary and draws some broad conclusions.
I I . T H E O RY A N D E M P I R I C A L A N A LY S I S
Policy-makers in developing economies have often set a broader agenda for

privatisation than the efficiency and resource allocation objectives that were
implicit in the policy conditions of structural adjustment programmes. The
motives for privatisation have encompassed improved fiscal equity and
distributional performance, although the importance attached to each has
varied between and within countries over time [UNCTAD, 1996; Yarrow,
1999]. Nevertheless, the link between privatisation and economic growth
relates most directly to the microeconomic theories used to justify
privatisation. At the heart of the debate are theoretical perspectives on the
ownership issue drawn from property rights theory, public choice theory
and principal agent analysis [Alchian, 1965; Tullock, 1965; Jensen and
Meckling, 1976]. By the end of the 1970s these theories were influencing
attitudes towards public ownership among policy-makers in developed and
developing countries [Cook and Kirkpatrick, 1988; Martin and Parker,
1997]. The key theoretical elements underpinning the argument for a
change in ownership from public to private related, first, to the view that
396jds05.qxd 29/08/03 14:13 Page 123
P R I VAT I S AT I O N A N D E C O N O M I C G R O W T H 123
public ownership led to the pursuit of objectives that detracted from

economic welfare maximisation [Boycko, Shleifer and Vishny, 1996].
Secondly, an ownership change could improve economic performance by
changing the mechanisms through which different institutional
arrangements affect the incentives for managing enterprises [Vickers and
Yarrow, 1988; Laffont and Tirole, 1991; Cook and Fabella, 2002].
These arguments are linked to presumptions concerning the condition of
publicly-owned enterprises before they are privatised. A typical view
presented publicly-owned enterprises as overextended and poor performers
[Kikeri, Nellis and Shirley, 1994]. In this situation publicly-owned
enterprises crowded out private enterprises in their access to credit and
erected statutory barriers to preserve the monopoly status of publicly-owned
enterprises. It was argued that the net effect of a change in ownership from
public to private would be improved economic efficiency, and over time an
increase in investment.
If privatisation was sufficiently extensive and had efficiency-inducing
effects, the contribution of improved performance could be detected at the
macroeconomic level. Privatisation would reduce crowding out and provide
more credit to the private sector. It would increase the opportunities for
investment in newly privatised enterprises by releasing them from the
capital constraints previously faced under public ownership. A change in
ownership would increase efficiency by introducing changes to the
governance mechanisms and structure of incentives facing employees.
Attempting to measure the contribution of an ownership change on
economic growth is complicated by the fact that economic performance is
likely to be affected by factors that affect the wider economic environment
in which privatised enterprises operate. Privatisation is often accompanied
in developing countries by changes in economic policies that affect
economic growth. Significant attention has focussed on the process of
deregulation and the importance of competition and its relation to economic
efficiency. Vickers and Yarrow [1988] have argued that competition and its
regulation are crucial for the improvement of efficiency in privatised
enterprises. Unravelling the separate effects of policy changes and degrees
of competition is difficult, and partly explains the relative deficiency of
empirical analysis in this area. The other major constraint to the
development of empirical investigations has obviously related to the time
period that has elapsed since privatisation. Until recently insufficient data
was available to carry out studies capable of measuring the dynamic effects
of privatisation.
Two recent studies have attempted to measure the impact of privatisation
on economic growth in developing countries. The first, by Plane [1997],
was used for 35 developing countries covering the period 1984–92. The
396jds05.qxd 29/08/03 14:13 Page 124

study, using Probit and Tobit models, found that privatisation positively
affected GDP growth and that the effect on growth was more significant for
activities of a public goods type than for other sectors. The study concludes
that on average institutional reform increased economic growth from 0.8 per
cent to 1.5 per cent between the sub-periods 1984–88 and 1988–92.
The second study was published by the IMF in 2000 [Barnett, 2000].
This study investigated the effect of privatisation on real GDP growth,
unemployment, and investment. Only twelve developing countries were
used in this 18-country study. The rest consisted of transitional economies.
The empirical analysis from this study strongly supported the hypothesis
that privatisation was positively correlated with real GDP growth. They
found that privatisation of a one per cent of GDP was associated with an
increase in the real growth rate of 0.5 per cent in period one and 0.4 per cent
in period two. These periods vary for each country to reflect periods of
active privatisation, but the precise span of years for the study is not
specified. It has also been suggested that the privatisation variable used in
Barnett’s is likely to capture the positive impact of a general regime change
towards better economic policies [Davis, Ossowski, Richardson and
Barnett, 2000].
I I I . M E T H O D O L O G Y A N D D ATA
In most developing countries privatisation has been a new policy instrument

and has typically been implemented sporadically rather than annually. As a
result, this study applies a cross-country regression analysis to examine
whether or not privatisation affects economic growth. Cross-country
regression analysis has been widely used to scrutinise the determinants of
long-run growth. These studies have identified various policy and socio-
demographic factors that may contribute to long-run growth. Studies of
these types and their interpretation have led to debates over the validity of
these approaches.
It has become apparent that cross-country growth regressions potentially
suffer a number of statistical and conceptual problems. Harberger [1987]
points out that one of the most fundamental statistical problems could be
caused by the inclusion of countries that have little in common into the same
regression. In principle, regression analysis necessitates observations that
are drawn from a distinct population. Thus, the inclusion of the countries
which are intrinsically different may result in unacceptable levels of
statistical bias [Levine and Zervos, 1993].
An important conceptual problem relates to the way in which cross-
country growth regression analysis is conducted. A typical cross-country
growth regression analysis uses data that are averaged over a significantly
396jds05.qxd 29/08/03 14:13 Page 125
long period, to permit the coefficients on various variables to be interpreted

as elasticities. In turn, these coefficients are used to indicate the percentage
growth that results from a one percent change in a policy variable. None the
less, Levine and Zervos [1993] warn that this type of analysis ought to be
treated cautiously since a cross-country regression itself does not provide
the complete answer to any causal relationships between policy variables
and growth. Accordingly, they argue that ‘cross-country regressions should
be viewed as evaluating the strength of partial correlations, and not as
behavioural relationships that suggest how much growth will change when
policies change’ [1993: 426].
In practice cross-country regression analysis has been one of the most
widely used methods in research on the relationship between policy changes
and growth. In order to take advantage of its usefulness, it is imperative to
minimise any possibility that distorts results, and to check for the robustness
of results. With these reservations in mind, this paper carries out a cross-
country growth regression analysis using the framework of the so-called
extreme-bounds analysis (EBA). The EBA was developed by Leamer
[1978, 1983, 1985] and Leamer and Herman [1983] and has been applied
practically in a number of empirical studies [Cooley and Leroy, 1981; Dervi
et al., 1990; Levine and Renelt, 1992; Levine and Zervos, 1993].
The EBA can be conducted by applying the following linear ordinary
least squares (OLS) regression:
Y = β1I + β2M + β3Z + u,
where Y is the GDP per capita growth rate, I is a set of variables that are
commonly included in the regression, M is the policy variable of particular
interest, and Z is a set of up to three variables chosen from a pool of policy
variables. This method follows Levine and Renelt [1992] and is different
from the original EBA developed by Leamer [1978], which includes all the
Z variables in a regression. It is more practical and has been referred to as a
modified version of EBA [Doppelhofer, Miller, and Sala-i-Martin, 2000].
In accordance with the EBA procedure, the combinations of a set of the
Z variables are changed to examine whether or not the coefficient on M is
statistically significant and maintains the same sign throughout the process.
To conduct the EBA, all possible combinations of regressions are estimated
to obtain β2 and the corresponding standard deviation, σ, for each
regression. Then, the upper or maximum extreme bound is estimated as the
maximum value of β2 plus σ2. The lower or minimum extreme bound is
defined as the minimum value of β2 minus σ2. If these maximum and
minimum extreme bounds keep the same sign, the result is interpreted as
robust rather than ‘fragile’ (see Appendix 1 for a more detailed account of
the EBA).
396jds05.qxd 29/08/03 14:13 Page 126

Several potential problems relating to EBA have been identified [Sala-

i-Martin, 1994]. Significant among these is the so-called reverse data
mining problem where the control variables in a regression are drawn from
the true population. This implies that all data contain some element of error.
By seeking different combinations of control variables, it is highly likely
then that there exists a combination that results in a variable of interest
being statistically insignificant or changing its coefficient sign. As a result,
the EBA ‘may be too strong’ [Sala-i-Martin, 1994: 743].
Accordingly, Sala-i-Martin [1997] has attempted to estimate the entire
distribution of β2. He also suggests an alternative method to conduct a
sensitivity analysis, although the statistical rationale for the approach has
been shown to be uncertain [Folster and Henrekson, 2001]. More recently,
Doppelhofer, Miller, and Sala-i-Martin [2000] have developed an
alternative method based on a Bayesian approach. Our analysis, however,
continues to use the modified EBA developed by Levine and Renelt [1992]
as these later contributions are still in their experimental stage.
Privatisation Variable
The most crucial question is how to measure the magnitude of privatisation
in a given country. Plane [1997] has used the cumulative proceeds from
privatisation during the period 1988–92 as a share of GDP in 1990 and a
dummy variable based on the information obtained from this indicator. A
similar indicator is used in this study, although the average GDP during the
sample period is used as a weight instead of choosing a particular year. This
reduces the arbitrariness from choosing a particular year for which the
sample countries might experience a variety of external or internal shocks.
In addition, the use of a dummy variable for privatisation is dropped
since it cannot convey information on the magnitude of privatisation into
the regression.
The dataset for privatisation is summarised in Table 1. The availability
of privatisation data largely determines the number of sample countries and
the sample period. As a result, 63 developing countries and the period
1988–97 are selected. The dataset shows the highly concentrated pattern of
privatisation, despite its adoption in a wide range of countries. Latin
American countries dominate the table in terms of privatisation as a share
of GDP. The dataset and full list of other variables corresponding to these
countries are shown in Appendix 2.
To reduce the risk that the privatisation variable reflects other policy and
structural reforms at the macroeconomic or aggregate level, it is necessary
to check if the effect of this variable can be isolated. This can be achieved
by examining the relationship between some of the policy variables that
might be captured by the privatisation variable. The effects of the
Downloaded By: [Hiroshima University] At: 07:16 25 July 2008 396jds05.qxd
TA B LE 1
C U M U L AT I V E P R O C E E D S F R O M P R I VAT I S AT I O N ( C P P ) A N D AV E R A G E G D P ( A G D P ) D U R I N G T H E P E R I O D 1 9 8 8 – 9 7
(IN US$ MILLIONS)
Country CPP AGDP CPP/ Country CPP AGDP CPP/

AGDP (%) AGDP (%)
29/08/03
1 Argentina 27921.0 28267.3 9.9 33 Tanzania 140.8 15031.3 0.9

2 Peru 7477.5 87051.1 8.6 34 Uganda 151.6 16429.3 0.9
3 Malaysia 10076.3 118695.4 8.5 35 Zimbabwe 197.3 22240.3 0.9
4 Singapore 4578.0 59669.2 7.7 36 Honduras 93.5 11003.2 0.8
14:13
5 Belize 54.3 793.0 6.8 37 Kenya 235.1 28329.6 0.8

6 Jamaica 532.9 8122.0 6.6 38 Togo 38.8 5220.3 0.7
7 Panama 832.8 13363.5 6.2 39 India 7073.0 1172166.0 0.6
8 Trinidad & Tobago 448.4 7599.4 5.9 40 Korea, Republic 2546.0 442276.5 0.6
9 Zambia 417.2 7865.9 5.3 41 Guinea 45.0 9928.1 0.5
10 Mexico 33353.1 647900.5 5.1 42 Jordan 58.7 11959.8 0.5
Page 127
11 Papua New Guinea 223.6 4397.3 5.1 43 Thailand 1378.4 296925.5 0.5
12 Bolivia 884.2 17720.7 5.0 44 Tunisia 171.0 37380.8 0.5
13 Brazil 34559.4 877419.5 3.9 45 Oman 60.1 13370.7 0.4
14 Ghana 872.6 23098.0 3.8 46 Ecuador 169.4 48336.9 0.4
15 Venezuela 5914.1 168554.8 3.5 47 Mali 21.9 6015.3 0.4
16 Colombia 5685.1 215739.9 2.6 48 Costa Rica 56.7 18735.7 0.3
17 Barbados 51.0 2039.5 2.5 49 Paraguay 42.0 15767.2 0.3
18 Morocco 1846.7 77307.4 2.4 50 Cameroon 41.1 23779.9 0.2
19 South Africa 6064.4 257829.9 2.4 51 Guinea-Bissau 0.5 235.2 0.2
20 Cote d’Ivoire 476.3 20874.1 2.3 52 Malawi 10.8 5549.1 0.2
21 Sri Lanka 725.9 34640.0 2.1 53 Bangladesh 60.3 98712.6 0.1
22 Nicaragua 130.2 6520.6 2.0 54 Burkina Faso 6.4 8411.6 0.1
23 Egypt 2777.7 142925.9 1.9 55 Burundi 4.2 4332.2 0.1
24 Philippines 3810.0 204678.8 1.9 56 Guatemala 43.4 33999.6 0.1
25 Senegal 191.4 12344.0 1.6 57 Nepal 15.0 18530.0 0.1
26 Turkey 3843.4 306790.8 1.3 58 Sierra Leone 1.6 2396.3 0.1
27 Mozambique 110.6 8932.9 1.2 59 Uruguay 17.0 23925.7 0.1
28 Pakistan 1951.0 161154.2 1.2 60 Iran 18.1 209417.6 0.0
29 Chile 1484.1 129306.1 1.1 61 Mauritania 1.1 3357.2 0.0
30 Benin 56.5 5692.5 1.0 62 Vietnam 2.6 72856.0 0.0
31 Indonesia 5162.8 494440.0 1.0 63 Yemen 0.8 8544.5 0.0
32 Nigeria 763.5 87263.8 0.9
Source: Primary data from World Bank Privatization Database and Privatisation International (various issues), and World Bank Global Development
Network Growth Database.
396jds05.qxd 29/08/03 14:13 Page 128

privatisation variable may be linked to the size of the government budget

deficit, since privatisation may have been implemented as a quick solution
to a budgetary problem in developing countries [Yarrow, 1999]. In order to
rule out this possibility, a simple correlation between the privatisation
indicator and the average ratio of government budget deficit to GDP during
the sample period 1988–97 was examined.
The result, shown in Figure 1, indicates that there is no correlation
between the two variables (the correlation coefficient is 0.23). Argentina,
Peru, and Malaysia implemented sizeable privatisation programmes while
maintaining relatively low budget deficits. Guinea-Bissau and Mozambique
recorded considerably high budget deficits with relatively small
privatisation programmes. It should be emphasised that this simple
correlation analysis, conducted without lags, does not eliminate the
possibility that some countries might have reduced their budgetary deficits
as a result of privatisation.
Similarly, the privatisation variable may capture the effects of policy
reforms pursued through World Bank adjustment programmes. In our
sample, 56 of the 63 countries received various adjustment loans with
policy reform conditions during the period 1985–97. In particular,
Argentina, Bangladesh, Bolivia, Ghana, Jamaica, Mauritania, and Zambia
F I GURE 1
T HE CORRE L AT I ON BE T WE E N PRIVATISATIO N A N D
GOVE RNME NT BUDGET D EFIC IT
10
9
8
7
PRIV/GDP
6
5
4
3
2
1
0
-30 -25 -20 -15 -10 -5 0 5 10 15
BUD/GDP
Note: PRIV/GDP and BUD/GDP are the average percentages during the period 1988–97.
Primary data from World Bank Privatisation Database and World Bank Global
Development Network Growth Database.
396jds05.qxd 29/08/03 14:13 Page 129
received adjustment loans with explicit conditions to implement

privatisation programmes.
The World Bank states that ‘privatisation became an important feature in
the Bank’s adjustment lending from the mid-1980s. This reflected both the
concurrent ideological shift in favour of private ownership as well as the
Bank’s long, and painful, experience with failed attempts at reforming
public enterprises’ [World Bank, 1995b: 56]. Before 1985, eight per cent of
all public enterprise reform-related adjustment loans attached conditions
regarding the implementation of privatisation. Between 1985 and 1989, the
proportion increased to 14 per cent and to 32 per cent in the period between
1990 and 1994 [World Bank, 1994].
A simple correlation was conducted between privatisation and World
Bank adjustment loans in order to eliminate the possibility that the
privatisation variable captures the macroeconomic effects of adjustment
programmes. The magnitude of World Bank adjustment loans is calculated
in the same manner as the privatisation indicator (i.e. the cumulative amount
of World Bank adjustment loans during the period 1985–97 divided by the
average GDP during the same period). The result reported in Figure 2
clearly indicates that there is no apparent relationship between privatisation
and adjustment loans (the correlation coefficient is 0.03).
F I GURE 2
T HE CORRE L AT I ON BE T WE E N PRIVATISATIO N A N D
WORLD BANK ADJUSTMENT LOANS
12
10
8
PRIV/GDP
0
0 2 4 6 8 10 12 14 16
SAL/GDP
Note: PRIV/GDP and SAL/GDP are the average percentage during the period 1988–97. Primary
data from World Bank Privatisation Database, World Bank Global Development Network
Growth Database, and World Bank SAL Database.
396jds05.qxd 29/08/03 14:13 Page 130

Of course, these results do not automatically confirm that the

privatisation indicator is able to capture the effects of privatisation at the
macroeconomic level. It only assures us that the privatisation indicator is
not picking up the effects of other policy reforms. In addition, since our
study has significantly increased the number of sample countries compared
with the previous studies, then the possibility that the privatisation indicator
captures the effects of other macroeconomic reforms is considerably
reduced.
IV. EXTREME BOUNDS ANALYSIS (EBA )
The analysis is conducted in two parts. In the first, the control variables are
examined to find appropriate specifications for the I variables. In the second
part, the results of introducing privatisation are reported.
The Basic Test for the Control Variables (the I variables)

In principle, the I variables are always included in an EBA. These variables
have to be selected with great care. In the context of a cross-country growth
regression, the control variables ought to control for initial economic,
political, and social conditions of countries. There are several variables that
are routinely included in these types of regressions. They are: (1) log of
initial GDP per capita; (2) initial life expectancy at birth; (3) average
population growth rate; (4) the ratio of government consumption to GDP;
(5) the ratio of gross domestic investment to GDP; and (6) the rate of
secondary school enrolment or attainment. These variables are sometimes
called ‘Barro-regressors’ and are used widely in cross-country growth
regressions [Barro, 1991].
Earlier, Levine and Zervos [1993] have argued that the inclusion of a
fiscal variable (4) does not necessarily control for initial conditions or
political stability but is a contemporaneous economic policy indicator.
However, Barro and Sala-i-Martin [1995] have argued that a fiscal variable
can be used as a proxy ‘for political corruption or other aspects of bad
government, as well as for the direct effects of non-productive public
expenditures and taxation’ [1995: 434].
The inclusion of investment as a control variable also remains
controversial. Levine and Renelt [1992] have included the ratio of
investment to GDP in their I variables, although a number of studies have
shown that causality runs from growth to investment and not vice versa
[Blomstrom, Lipsey and Zejan, 1993; Barro and Sala-i-Martin, 1995]. More
recently, Temple [1999] argues that the correlation is reverse and robust.
Besides the concern over the direction of causality, there is also a question
over what measure of investment to use. In general, gross domestic
396jds05.qxd 29/08/03 14:13 Page 131
investment is often used as a proxy for investment. However, Barro and

Sala-i-Martin [1995] argue that since gross domestic investment includes
both private and public spending, it may be an inappropriate measure of
investment. DeLong and Summers [1991] also argue that the use of gross
domestic investment can be ruled out since producer durables are the
significant component of investment. Since we are unable to resolve the
causality problem, we test for different combinations of the I variables using
both gross domestic investment and government consumption.
Another specification problem we have encountered relates to the
inclusion of the secondary school enrolment ratio. This was not included in
our study since it does not capture the quality of education. Barro and Sala-
i-Martin [1995] have argued that it is more appropriate to use the secondary
school attainment ratio, although if we use this variable, the number of
countries will have to be severely reduced owing to a lack of data. Instead,
the initial life expectancy at birth is used. This variable captures various
aspects of human capital development, such as investment in education and
increases in schooling, thereby reflecting the quality of education in general
[Ram and Schulz, 1979; Sala-i-Martin, 1997; Kalemili-Ozcan, Ryder and
Weil, 2000].
Combinations of variables were used to find the I variables relevant for
this study. This was achieved by examining four different combinations of
selected I variables. The first specification included the log of initial GDP
per capita for 1988 (hereafter LGYP), the initial life expectancy at birth for
1987 (LIFE), the average population growth during the sample period
1988–97 (POP), and the average ratio of GDI to GDP during the sample
period (GDI). The second specification included LGYP, LIFE, POP, GDI,
and the average ratio of government consumption to GDP during the sample
period (GOVC). The third specification included LGYP, POP, LIFE, and
GOVC. Finally, the fourth specification consisted of LGYP, LIFE, and POP.
In all cases the dependent variable is the average growth rate of GDP per
capita during the sample period (GYP).
In order to check the robustness of the basic results, a series of EBAs
was conducted by including the Z variables. The following Z variables were
selected: (1) openness, which is calculated as (export + import)/GDP
(hereafter OPEN); (2) informal market premium (IMP); (3) the standard
deviation of IMP (SDIMP); (4) the average ratio of FDI to GDP (FDI); (5)
the average number of major political crises, revolutions, and coups as a
proxy for political instability (STAB); (6) the average inflation rate (INFL);
(7) the standard deviation of average inflation rates (SDINF); (8) the
average ratio of liquid liabilities (M3) to GDP used as a proxy for the level
of financial sector development (M3); (9) the standard deviation of M3
(SDM3); (10) the average ratio of total external debt to GDP (DEBT); (11)
396jds05.qxd 29/08/03 14:13 Page 132

the average ratio of government budget surplus/deficit to GDP (BUD); and

(12) three regional dummies for East Asia, Latin America and Sub-Saharan
Africa (EA, LA, AF, respectively). Some of the I variables are also included
as Z variables. They are GOVC (in the first and fourth specifications) and
GDI (in the third and fourth specifications).
These variables, with the exception of (4), are frequently used in cross-
country growth regressions. The ratio of FDI to GDP is included on the
assumption that FDI may play a significant role in generating positive
spillover effects, bringing new technologies and management and
marketing skills that contribute to economic growth in developing
countries. In addition, STAB is included as a Z variable, although it could
be used as a control variable since there may exist a reverse causality that
runs from political instability to economic growth. In this respect,
Londregan and Poole [1990] argue that higher political instability is likely
to be caused by lower economic growth, and Barro [1991] and Alesina and
Petrotii [1993] have also found evidence supporting this reverse causality.
The EBAs for the control variables in the four specifications were
conducted by computing a total of 7,140 regressions, and the results are
shown in Table 2. Note that any result obtained from a regression that
exhibits a significant level of multicollinearity (that is, a regression showing
that a variance inflation factor (VIF) is greater than 10) was discarded from
these results.
In general, the inclusion of LGYP is expected to represent the rate of
convergence across the sample countries. Other studies that have included
developed countries have found evidence of conditional convergence [e.g.,
Mankiw, Romer and Weil, 1992; Barro and Sala-i-Martin, 1995]. A
negative and statistically significant coefficient for this variable signifies
evidence of conditional convergence. In our study, the results for LGYP
obtained from the first, third, and fourth specifications appear to be negative
and fragile (see the notes in Table 2 for the criterion used to interpret EBA
results).
With respect to the results for the first and second specifications, LIFE
appears to be fragile, although robust results were obtained in the third and
fourth specifications. Throughout, the sign of the coefficient is positive, and
the robust results imply that this variable captures aspects of human capital
development reasonably well. A robust negative result was obtained for
POP from the first and forth specifications. In contrast, the result obtained
from the second specification was completely insignificant and was fragile
for the third. These contrasting results may suggest that this variable is
extremely sensitive to the inclusion or exclusion of other variables, as
indicated by Levine and Renelt [1992]. Robust results were obtained in
relation to GDI and GOVC. The signs of the coefficients for GDI are
396jds05.qxd 29/08/03 14:13 Page 133
TABL E 2
T HE E BA RE S ULT S OF T HE I VARIA BLES (AT 0.05 LEV EL)
Specification I
Variable Coefficient t-statistic Z-variables included EBA result
LGYP Min –0.012 –2.05 STAB, EA, AF Fragile

Max –0.011 –2.02 STAB, FDI, EA
Base –0.005 –0.94 –
LIFE Min 0.052 2.01 GOVC, SDIMP, LA Fragile
Max 0.061 2.23 GOVC, BUD, LA
Base 0.025 0.98 –
POP Min –1.065 –3.25 FDI, IMP, LA Robust
Max –0.650 –2.01 INFL, M3, EA
Base –0.782** –2.34 –
GDI Min 0.104 2.17 FDI, EA, LA Robust
Max 0.219 5.20 IMF, SDM3, DEBT
Base 0.203*** 5.03 –
Specification II
LGYP Min – – – Insignificant

Max – – –
Base –0.003 –0.56 –
LIFE Min 0.051 2.02 FDI, IMP, LA Fragile
Max 0.066 2.37 STAB, BUD, LA
Base 0.028 1.12 –
POP Min – – – Insignificant
Max – – –
Base –0.390 –1.07 –
GDI Min 0.108 2.25 STAB, EA, LA Robust
Max 0.209 5.03 BUD, SDM3, DEBT
Base 0.196*** 5.03 –
GOVC Min –0.160 –2.98 BUD, SDIMP, LA Robust
Max –0.107 –2.02 STAB, IMP, EA
Base –0.120** –2.29 –
Specification III

Max –0.011 –2.04 STAB, SDIMP, EA
Base –0.005 –0.82 –
LIFE Min 0.052 2.01 GDI, SDIMP, LA Robust

Max 0.109 3.95 STAB, BUD, LA
Base 0.068** 2.43 –
POP Min –0.760 –2.04 STAB, FDI, LA Fragile
Max –0.742 –2.01 FDI, SDIMP, LA
Base –0.354 –0.82 –
396jds05.qxd 29/08/03 14:13 Page 134

TABL E 2 (cont.)
Specification III (cont.)
GOVC Min –0.203 –3.17 STAB, BUD, M3 Robust

Max –0.109 –2.04 GDI, SDINF, SDM3
Base –0.139** –2.24 –
Specification IV

Max –0.011 –2.01 MDM3, EA, AF
Base –0.008 –1.20 –
LIFE Min 0.054 2.05 GOVC, FDI, M3 Robust
Max 0.103 3.77 GOVC, BUD, LA
Base 0.067** 2.30 –
POP Min –1.216 –3.46 FDI, SDM3, LA Robust

Max –0.660 –2.01 GDI, M3, EA
Base –0.810** –2.04 –
Notes: Dependent variable: GYP. The base results are calculated excluding the M and Z
variables. Note that following Levine and Renelt [1992] (see Table 1, notes (a), p.947),
where an EBA indicates a robust result, but the base result is statistically insignificant, the
overall result is interpreted as fragile. ** = statistically significant at 0.05 level, and ***
= statistically significant at 0.01 level.
positive, whereas they are negative for GOVC, confirming previous studies
[e.g., Barro, 1991].
In summary, LGYP exhibits fragile results, and LIFE and POP appear to
be sensitive to different combinations of other variables, resulting in
ambiguous results. On the other hand, the results for GDI and GOVC are
always robust. Nevertheless, it is difficult to choose a particular
specification for the control variables from these results. Since a continued
search for the most suitable combination of the control variables is beyond
the scope of this study and may prove impossible, as some past studies have
demonstrated, then all four specifications are used to conduct an EBA for
the privatisation variable.
The EBA for Privatisation

The EBA proceeded in two steps. First, basic tests were carried out to
estimate baselines before a formal EBA was conducted for privatisation.
These regressions included the control variables in the four specifications
396jds05.qxd 29/08/03 14:13 Page 135
examined previously and the privatisation variable. The Z variables were

excluded. The tests showed that the base results for privatisation were
statistically insignificant. Second, the EBA for the privatisation variable
was conducted by this time including three regressors from a pool of the 15
Z variables. The results are shown in Table 3. Although the EBA indicated
that the results were robust, they still have to be interpreted as fragile since
the earlier basic tests indicated that they were statistically insignificant.
Accordingly, at this stage, we have concluded that only a weak or fragile
negative partial correlation between privatisation and economic growth
exists.
TABL E 3
T H E E BA RE S ULT S ( AT 0.05 L E VE L ) F OR TH E PR IVATISATIO N VA R IA BLE
The I variables based on Specification I (LGYP + LIFE + POP + GDI)
PRIV Min –0.243 –2.25 BUD, FDI, DEBT Fragile

Max –0.219 –2.01 GOVC, FDI, IMP
Base –0.056 –0.55 – –
The I variables based on Specification II (LGYP + LIFE + POP + GDI + GOVC)
PRIV Min –0.249 –2.35 BUD, FDI, DEBT Fragile

Max –0.216 –2.04 BUD, FDI, EA
Base –0.077 –0.77 – –
The I variables based on Specification III (LGYP + LIFE + POP + GOVC)
PRIV Min –0.347 –2.82 FDI, SDM3, DEBT Fragile

Max –0.219 –2.01 GDI, FDI, IMP
Base –0.090 –0.76 – –
The I variables based on Specification IV (LGYP + LIFE + POP)
PRIV Min –0.345 –2.76 FDI, SDM3, DEBT Fragile

Max –0.228 –2.02 FDI, SDIMP, LA
Base –0.067 –0.55 – –
Notes: Dependent variable: GYP. Note that following Levine and Renelt [1992] (see Table 1,
notes (a), p.947), in the case that the EBA indicates a robust result, but the base result is
statistically insignificant, the overall result is interpreted as fragile.
396jds05.qxd 29/08/03 14:13 Page 136

Two cautionary notes are required at this point. First, it is known that the
inclusion or exclusion of certain control variables sometimes eliminates a
bivariate relationship [Easterly and Rebelo, 1993]. It is thus plausible to
argue that our inconclusive results are due to an inadequate choice or
combinations of the control variables. In order to examine this possibility,
all possible combinations of the control variables and the privatisation
variable were scrutinised by applying the EBA technique. This exercise
reconfirmed our conclusion by showing that no combination of control
variables exists that would establish a robust partial correlation between
privatisation and economic growth.
Second, it is known that cross-country regression analysis is sensitive to
the inclusion or exclusion of a particular country. This may arise due to the
existence of influential observations and/or outliers in the regressions. This
possibility is also examined by checking whether or not there are any
influential observations or outliers in our basic regression models. It should,
however, be pointed out that there is no universally accepted indicator that
is able to identify influential observations and outliers. For instance, Donald
and Maddala [1993] argue that the examination of studentised residuals is
the most appropriate method to identify influential observations, and to
detect outliers leverage should also be examined. On the other hand,
Beckman and Cook [1983] and Fiebig [1992] argue that large studentised
residuals may not necessarily point to influential observations, and Fiebig
[1992] specifically points out that influential observations may actually be
related to small residuals.
In addition, even if influential observations and outliers are identified, it
may prove difficult to deal with them. Influential observations and outliers
may arise as a result of data errors and model misspecifications. Suitable
corrections may be hampered because of poor data, especially for
developing countries, and simply deleting the countries in question may still
be appropriate. On the other hand, influential observations may represent
natural variation which provides valuable information, and eliminating
these observations would clearly be inappropriate. Even if influential
observations are kept in the regression, however, the potential effect on the
results of eliminating them ought to be noted [Fiebig, 1992]. One way to
accommodate influential observations in the regression is to apply robust
regression estimators instead of OLS (see Rousseeuw and Leroy [1987] for
a detailed account of robust regression estimators). It should also be noted
that even these detection techniques and robust estimators may fail to detect
outliers [Luke and Schaffer, 2000].
Accepting these limitations, studentised residuals were examined in an
attempt to detect influential observations in our basic regression models. We
also applied some of the influential observation and outlier detection
396jds05.qxd 29/08/03 14:13 Page 137
techniques, DFITS [Welsch and Kuh, 1977], Cook’s Distance [Cook, 1977],
and Welsch Distance [Welsch, 1982]. These techniques, in essence,
encapsulate the information in the leverage versus residual-squared plot into
a single statistic, creating an index that is influenced by the size of the
residuals or outliers and the size of leverage (see Appendix 3 for the
calculation procedures). In addition, we tested for influential observations
and outliers specifically to examine the privatisation variable by applying
DFBETAs [Belsley, Kuh and Welsch, 1980]. The application of DFBETAs
permits us to examine the difference between the coefficients of a specific
variable when a particular observation is included or excluded
Table 4 reports the results of the tests for influential observations.
Although it is difficult to compare the indices across the different
specifications and techniques, these tests identified 15 countries as potential
outliers or a total of 95 observations as influential. Singapore and Malaysia
contributed significantly to the influential observations, 17.9 per cent and
14.7 per cent respectively, followed by Oman 13.7 per cent, and to a lesser
extent Sierra Leone 8.4 per cent. These four countries generated more than
50 percent of the influential observations. In particular, Welsch Distance
indicates that only Malaysia, Oman, and Singapore are potential outliers.
More specifically, of those 20 test results shown in Table 4, Singapore
appears as a potential outlier in 17 of them, Malaysia appears in 14 tests,
and Sierra Leone is identified in eight tests. Oman was detected as
influential by 13 test results only in the first, second, and third
specifications. In addition, the combination of Malaysia and Singapore
appear in 14 test results, including all the test results in the third and fourth
specifications. Furthermore, the results of DFBETAs suggest that these two
countries have substantial positive effects on the privatisation variable, in
particular in the third and fourth specifications.
After identifying the influential observations and thus the potential
outliers in our basic regression models, we attempted to re-estimate the
baselines for privatisation in two steps. First, we re-estimated the baselines
for privatisation by applying two robust regression estimators, the median
least squares (MLS) and a robust regression estimator using iteratively
reweighted least squares, including all of the influential observations (see
Rousseeuw and Leroy [1986] and Luke and Schaffer [2000] for details of
these estimators). We then re-estimated the baselines using OLS by
eliminating all the influential observations. Second, we deleted the
influential observations one by one according to the higher absolute value
of each of the indicators for influential observations. Following these steps,
we could not find any statistically significant results for privatisation in the
first and second specifications but found a number of statistically
significant results for privatisation in the remaining specifications.
396jds05.qxd 29/08/03 14:13 Page 138

TABL E 4
T E S T S F OR I NF L UE NT I AL OBS E RVATIO N S A N D O U TLIERS
Specification I
Studentized DFBETAs DFITS Cook’s Welsch
1 Uganda 2.563 Jamaica –0.543 Oman 1.065 Jordan 0.305 Oman 10.09
2 Jamaica –2.559 Malaysia 0.437 Singapore 0.862 Sierra Leone 0.231 Singapore 7.648
3 Cameroon –2.514 Singapore 0.364 Vietnam 0.789 Jamaica 0.194
4 Jordan –2.341 Peru –0.356 Uganda 0.709 Oman 0.184
5 Sierra Leone –2.041 Oman –0.290 Malaysia 0.672 Singapore 0.120
6 Jordan 0.255 Vietnam 0.099
7 Uganda 0.077
8 Malaysia 0.075
Specification II
1 Cameroon –2.959 Jamaica –0.510 Oman 2.327 Oman 0.687 Oman 23.69
2 Oman 2.837 Oman –0.459 Singapore 0.742 Sierra Leone 0.247
3 Uganda 2.321 Peru –0.456 Vietnam 0.733 Jordan 0.235
4 Sierra Leone –2.287 Malaysia 0.409 Uganda 0.727 Jamaica 0.156
5 Jamaica –2.196 Singapore 0.279 Singapore 0.077
6 Jordan –2.148 Cameroon 0.254 Vietnam 0.074
7 Uganda 0.070
8
Specification III
1 Cameroon –2.810 Malaysia 0.756 Oman 1.327 Sierra Leone 0.323 Oman 13.38
2 Sierra Leone –2.549 Singapore 0.443 Singapore 1.048 Oman 0.285 Singapore 9.074
3 Malaysia 2.324 Peru –0.434 Malaysia 0.971 Singapore 0.170 Malaysia 8.289
4 Singapore 2.296 Korea –0.366 Korea 0.641 Malaysia 0.146
5 Thailand –0.302 Barbados 0.066
6 Argentina –0.294 Korea 0.065
7 Oman –0.260
Specification IV
1 Singapore 2.616 Malaysia 0.794 Singapore 1.081 Sierra Leone 0.326 Singapore 9.213
2 Cameroon –2.411 Singapore 0.537 Malaysia 0.987 Singapore 0.212 Malaysia 8.402
3 Malaysia 2.398 Korea –0.366 Korea 0.646 Malaysia 0.180
4 Sierra Leone –2.325 Peru –0.340 Barbados 0.100
5 Thailand –0.303 Korea 0.080
6 Jamaica –0.295 Jamaica 0.068
7 Jordan 0.065
396jds05.qxd 29/08/03 14:13 Page 139
Accordingly, the first and second specifications were dropped from any
further examination.
Correspondingly, we re-estimated the baselines for privatisation by
applying the two robust regression estimation techniques which enabled us
to incorporate the influence of the influential observations or potential
outliers in the regressions without deleting them. The results are reported in
Table 5 in the third and fourth columns of each specification. In the third
specification, we found statistically significant results for privatisation from
MLS (at the 0.10 level) and Robust Regressions (at the 0.01 level). In the
case of the fourth specification, we could not find statistically significant
results for privatisation from the MLS but found the statistically significant
results from Robust Regression (at 0.01 level). As for the MLS results from
specification four, pseudo R2 is too low, and the signs of the coefficients,
except those for PRIV and POP, changed. In essence, we have found that a
robust partial correlation between privatisation and economic growth can be
established in these specifications, in particular in the third specification, by
using the robust estimators instead of OLS and without deleting the
influential observations.
The results of excluding all the influential observations using OLS (a
total of 10 models were estimated) are shown from the fifth to the last
column in Table 5. In these specifications, we found statistically significant
results for privatisation at the 0.01 level (shown in column (1) in the third
specification and column (6) in the fourth specification) by deleting all the
influential observations identified by studentised residuals. Similarly, we
found statistically significant results for privatisation in the fourth
specification (column 10) at the 0.05 level and in the third specification
(column 5) at the 0.10 level by eliminating all the influential observations
identified by Welsch Distance. As for Cook’s Distance, we also found
statistically significant results for privatisation at the 0.05 level (the third
specification (column 4)) and at the 0.10 level (the fourth specification
(column 9)).
In contrast, the elimination of all the influential observations identified
by DFBETAs resulted in statistically insignificant results for privatisation in
both specifications. In terms of the deleted influential observations
identified by DFITS, the results are somewhat mixed. We found a
statistically significant result for privatisation at the 0.10 level in the fourth
specification (column 8) and statistically insignificant result in the third
specification (column 3). In sum, we tested ten models using OLS in the
third and fourth specifications and found statistically significant results in
seven models.
The results of eliminating the influential observations to examine
whether or not these have led to increases in R2 and/or the statistical
TA B L E 5
T H E B A S E L I N E R E S U LT S F O R P R I VAT I S AT I O N
Specification III
396jds05.qxd
OLS MLS Robust Regression (1) Studentised (2) DFBETAs (3) DFITS (4) Cook’s (5) Welsch
PRIV –0.090 –0.192* –0.254*** –0.292*** –0.063 –0.178 –0.214** –0.223*

(–0.76) (–1.89) (–2.94) (–2.91) (–0.46) (–1.58) (–2.01) (–1.96)
LGYP –0.004 –0.010* –0.006 –0.002 –0.016** –0.016** –0.009 –0.013**
(–0.64) (–1.72) (–1.34) (–0.37) (–2.66) (–2.56) (–1.50) (–2.18)
29/08/03
LIFE 0.069** 0.068*** 0.032 0.036 0.085*** 0.089*** 0.051* 0.085***

(2.46) (2.71) (1.55) (1.44) (3.57) (3.63) (1.95) (3.37)
POP –0.372 –0.648 –0.886*** –0.727** –0.749* –0.897** –1.156*** –1.012**
(–0.85) (–1.64) (–2.79) (–2.06) (–1.87) (–2.22) (–2.85) (–2.45)
GOVC –0.143** –0.133** –0.140*** –0.130** –0.156** –0.145** –0.129** –0.149**
14:13
(–2.28) (–2.44) (–3.08) (–2.60) (–2.50) (–2.48) (–2.25) (–2.48)

Constant –0.207** –0.156** –0.025 –0.077 –0.179** –0.191** –0.079 –0.186**
(–2.40) (–1.99) (–0.40) (–1.01) (–2.48) (–2.57) (–1.00) (–2.43)
Adjusted R 2 0.22 – – 0.31 0.30 0.32 0.31 0.34
Pseudo R2 – 0.13 – – – – – –
No. of influential – – – All (4) All (7) All (4) All (6) All (3)
obs. excluded
Page 140
Specification IV
OLS MLS Robust Regression (6) Studentised (7) DFBETAs (8) DFITS (9) Cook’s (10) Welsch
PRIV –0.067 –0.106 –0.250*** –0.285*** –0.094 –0.204* –0.221* –0.249**

(–0.55) (–1.24) (–2.61) (–2.71) (–0.74) (–1.73) (–1.95) (–2.10)
LGYP –0.007 0.002 –0.005 –0.006 –0.016*** –0.014** –0.008 –0.012*
(–1.06) (0.32) (–1.05) (–0.99) (–2.67) (–2.31) (–1.27) (–1.95)
LIFE 0.067** –0.011 0.023 0.038 0.085*** 0.082*** 0.051* 0.078***
(2.31) (–0.54) (0.99) (1.44) (3.37) (3.19) (1.83) (2.97)
POP –0.834** –1.340*** –1.185*** –1.171*** –0.934** –0.975*** –1.283*** –1.112***
(–2.08) (–4.69) (–3.79) (–3.60) (–2.60) (–2.69) (–3.15) (–3.04)
Constant –0.188** 0.083 –0.005 –0.066 –0.193** –0.193** –0.100 –0.187**
(–2.11) (1.31) (–0.08) (–0.83) (–2.57) (–2.49) (–1.18) (–2.35)
Adjusted R2 0.17 – – 0.24 0.23 0.23 0.22 0.25
Pseudo R2 – 0.09 – – – – – –
No. of influential – – – All (4) All (6) All (3) All (7) All (2)
obs. excluded
Notes: Dependent variable: GYP. The models (1) - (10) were estimated using OLS. * = statistically significant at 0.10 level , ** = statistically significant at
0.05 level, and *** = statistically significant at 0.01 level. t-statistics are in parentheses.
396jds05.qxd 29/08/03 14:13 Page 141
significance of the privatisation variable are shown in Table 6. Statistically

significant results were obtained for privatisation in all six models estimated
in the third and fourth specifications. In these models, higher adjusted R2
and/or increased statistical significance for privatisation were obtained by
deleting a few countries from a pool consisting of Barbados, Cameroon,
Malaysia, Oman, Sierra Leone and Singapore. Malaysia and Singapore
appear to be the most influential observations. In two models (column 11)
and (column 16), higher R2 and increased statistical significance for
privatisation were obtained by eliminating these two countries, and in the
remaining 4 models (shown in columns 12, 13, 14 and 15), higher R2 and
increased statistical significance were obtained by eliminating a few of the
countries including Malaysia and Singapore. Accordingly, Malaysia and
Singapore appear to bias our regression results (see Stokey [1994] and
Lorgelly and Owen [1999] for a discussion of biased regression results
caused by the inclusion of certain East Asian countries).
In summary, we obtained a total of 14 new estimates for privatisation,
using OLS and deleting the influential observations identified by five
different outlier detection techniques (note that some of the results are
identical: (column 5) and (column 12) in the third specification and (column
10) and (column 16) in the fourth specification). In order to conduct the
EBAs for these new estimates, we have selected from Tables 5 and 6
baseline results for privatisation, which are statistically significant at 0.05
level, for each specification according to the highest adjusted R2. This was
done because the elimination of a subset of some influential observations
from our dataset resulted in the largest reduction in the residual sum of
squares. It is likely that the eliminated observations are outliers [Gentleman
and Wilk, 1975].
All the baseline results were adopted from Table 6. We chose the
baseline results obtained by Cook’s Distance (the third and fourth
specifications) and the studentised residuals (the fourth specification). The
most likely candidates for outliers turned out to be Barbados, Cameroon,
Oman, Malaysia, Sierra Leone, and Singapore. On this basis, Table 7 shows
EBAs for privatisation by deleting the influential observations or the most
likely outliers. This time, robust results were obtained from all two
specifications. We also conducted the EBAs for the other statistically
significant base results for privatisation at the 0.05 level by deleting the
influential observations in Tables 5 and 6. The EBA results for privatisation
are negative and robust in all the models.
Consequently, a negative but fragile partial correlation between
privatisation and economic growth has been found using the full 63-country
sample with OLS estimators, and a robust negative partial correlation has
been found by eliminating the possible outliers. We have also found a robust
396jds05.qxd 29/08/03 14:13 Page 142

TABL E 6
T H E B A S E L I N E R E S U LT S F O R P R I VAT I S AT I O N E X C L U D I N G
S E L E CT E D I NF L UE NT I AL O BSERVATIO N S
Specification III
Full sample (11) DFBETAs (12) DFITS (13) Cook’s
PRIV –0.090 –0.252** –0.223* –0.256**

(–0.76) (–2.16) (–1.96) (–2.40)
LGYP –0.004 –0.009 –0.013** –0.007
(–0.64) (–1.52) (–2.18) (–1.14)
LIFE 0.069** 0.078*** 0.085*** 0.047*
(2.46) (3.03) (3.37) (1.72)
POP –0.372 –0.761* –1.012** –1.290***
(–0.85) (–1.86) (–2.45) (–3.16)
GOVC –0.143** –0.103* –0.149** –0.128**
(–2.28) (–1.77) (–2.48) (–2.18)
Constant –0.207** –0.201** –0.186** –0.069
(–2.40) (–2.56) (–2.43) (–0.85)
Adjusted R2 0.22 0.28 0.34 0.34
Likely outliers – Malaysia, Oman, Sierra Leone,
excluded Singapore Singapore, Oman,
Malaysia Singapore,
Malaysia,
Barbados
Specification IV
Full sample (14) DFBETAs (15) DFITS (16) Cook’s
PRIV –0.067 –0.271** –0.288** –0.249**

(–0.55) (–2.41) (–2.61) (–2.10)
LGYP –0.007 –0.011* –0.005 –0.012*
(–1.06) (–1.95) (–0.77) (–1.95)
LIFE 0.067** 0.075*** 0.038* 0.078***
(2.31) (3.01) (1.40) (2.97)
POP –0.834** –1.063*** –1.414*** –1.112***
(–2.08) (–3.07) (–4.03) (–3.04)
Constant –0.188** –0.179** –0.069 –0.187**
(–2.11) (–2.38) (–0.84) (–2.35)
Adjusted R2 0.17 0.27 0.27 0.25
Likely outliers – Singapore, Sierra Leone, Singapore,
excluded Cameroon, Singapore, Malaysia
Malaysia Malaysia,
Barbados
Notes: Dependent variable: GYP. These models were estimated using OLS. * = statistically
significant at 0.10 level , ** = statistically significant at 0.05 level, and *** = statistically
significant at 0.01 level. . t-statistics are in parentheses. The likely outliers excluded are
shown in the order of higher absolute values.
396jds05.qxd 29/08/03 14:13 Page 143
TABL E 7
T H E E BA RE S ULT S ( AT 0.05 L E VE L ) F OR TH E PR IVATISATIO N VA R IA BLE
E XCL UDI NG OU TLIER S
Specification III (LGYP + LIFE + POP + GOVC)
Coeffi- t-sta- Z-variables included Likely outliers EBA result

cient tistics excluded
Cook’s (13) Min –0.357 –3.23 FDI, SDM3, DEBT Sierra Leone,
Max –0.209 –2.11 GDI, FDI, LA Oman, Robust
Base –0.256** –2.40 – Singapore,
Barbados
Specification IV (LGYP + LIFE + POP)
Coeffi- t-sta- Z-variables included Likely outliers EBA result

cient tistics excluded
Studentised Min –0.373 –3.21 OPEN, FDI, DEBT Singapore,

(14) Max –0.203 –2.01 GOVC, GDI, BUD Cameroon, Robust
Base –0.271** –2.41 – Malaysia
Cook’s (15) Min –0.383 –3.34 OPEN, FDI, DEBT Sierra Leone,
Max –0.214 –2.02 FDI, EA, LA Sinagpore, Robust
Base –0.288** –2.61 – Barbados
Notes: Dependent variable: GYP. The EBA results for PRIV are obtained by including the control
variables and three regressors from the pool of the Z variables. ** = statistically significant
at 0.05 level. The likely outliers excluded are in the order of higher absolute values.
negative partial correlation between these two variables using the full 63-
country sample with the robust regression estimators. These results indicate
that the privatisation variable is sensitive to combinations of control
variables and the inclusion of certain countries. In other words, in order to
detect a significant result for the privatisation variable, great care is required
in the selection of a specific conditioning information set and countries.
V. C O N C L U S I O N
The analysis has shown that there is a robust partial correlation between
privatisation and economic growth, suggesting that privatisation has
contributed negatively to economic growth. This conclusion is contrary to
the results obtained by Plane [1997] and Barnett [2000]. The analysis in this
study has used a different approach to that adopted in Plane’s study. By
using an extreme bound analysis and a larger sample of countries we have
been able to undertake a more comprehensive and rigorous analysis of the
396jds05.qxd 29/08/03 14:13 Page 144

relationship between privatisation and economic growth. The analysis has

also been strengthened by the application of outlier detection techniques.
Despite these improvements in methodology, the results obtained from this
type of analysis must continue to be treated with a fair degree of caution.
Are there policy implementations arising from this analysis? Clearly, the
findings do not refute the notion that the wider economic and socio-political
environment may have important effects on economic growth, and on the
success of privatisation. The IMF study, using a limited number of
countries, has suggested that the positive relationship between privatisation
and economic growth could be explained by privatisation acting as a proxy
in their analysis for a range of structural measures signifying changes in the
economic environment. We have attempted, with our measure of
privatisation, to rule out the possibility that this variable captures the effects
of wider policy reforms.
If, as is argued in the theoretical literature, the change in ownership alone
at the microeconomic level is not sufficient to guarantee greater enterprise
efficiency, then other reforms, more directly related to enterprise
development, may indeed play a crucial role. If the success of privatisation
is linked to competition and the regulation of competition, then weaknesses
in these fields may explain why privatisation is negatively related to
economic growth in developing countries. Recent reviews of competition
policy in developing countries indicate fundamental weaknesses in
implementation [Cook, 2001]. Similarly, regimes for regulating competition
in developing countries have not developed uniformly and with the same
degree of effectiveness across developing countries. The share of utilities in
privatisation has increased significantly in the 1990s in developing countries,
resulting in the creation of numerous private sector monopolies and the need
for better regulation. Even in countries in Latin America where regulatory
systems have developed they may not be working effectively [Cook, 1999].
In addition, the IMF report evidence of weak regulatory systems in the
telecommunications sector in some developing countries, which may have
contributed to enhancing the proceeds from privatisation [Davis, Ossowski,
Richardson and Barnett, 2000]. The combination of weak regulation and
high proceeds from privatisation, if accompanied by low gains in economic
efficiency, may also provide an explanation for the disappointing results as
far as economic growth is concerned.
The policy implications arising from this analysis squarely point to the
need for further research into the policy environment, in particular the roles
played by competition and its regulation if the relation between
privatisation and economic growth is to be better understood.
final revision accepted July 2002
396jds05.qxd 29/08/03 14:13 Page 145
REF ERENCES
Alchian, A., 1965, ‘Some Economics of Property Rights’, II Politico, Vol.30, pp.816–29.
Alesina, A. and R. Petrotii, 1993, ‘Income Distribution, Political Instability and Investment’,
NBER Working Paper, No.4486.
Barnett, S., 2000, ‘Evidence on the Fiscal and Macroeconomic Impact of Privatisation’, IMF
Working Paper, July, Washington, DC: IMF.
Barro, R., 1991, ‘Economic Growth in a Cross-Section of Countries’, Quarterly Journal of
Economics, Vol.106, No.2, pp.407–43.
Barro, R. and X. Sala-i-Martin, 1995, Economic Growth, New York: McGraw-Hill.
Beckman, R. and R. Cook, 1983, ‘Outliers’, Technometrics, Vol.25, pp.119–49.
Belsey, D., Kuh, E. and R. Welsch, 1980, Regression Diagnostics, New York: John Wiley.
Bennell, P., 1997, ‘Privatisation in Sub-Saharan Africa: Progress and Prospects during the
1990s’, World Development, Vol.25, No.11, pp.1785–803.
Blomstrom, M., Lipsey, R. and M. Zejan, 1993, ‘Is Fixed Investment the Key to Economic
Growth?’ NBER Working Paper, No.4436.
Boubakri, N. and J. Cosset, 1998, ‘The Financial and Operating Performance of Newly Privatised
Firms: Evidence from Developing Countries’, unpublished draft, Universite Laval.
Boycko, M., Shleifer, A. and R. Vichny, 1996, ‘A Theory of Privatisation’, Economic Journal,
Vol.106, pp.309–19.
Cook, P., 1999, ‘Privatisation and Utility Regulation in Developing Countries: The Lessons So
Far’, Annals of Public and Cooperative Economics, Vol.70, No.4, pp.549–87.
Cook, P., 2001, ‘Competition and its Regulation: Key Issues’, paper presented at an international
workshop on ‘Competition and Regulation’, University of Manchester, 9 Feb. 2001
Cook, P. and C. Kirkpatrick, 1988, Privatisation in Less Developed Countries, Brighton:
Weatsheaf.
Cook, P. and R. Fabella, 2002, ‘The Welfare and Political Economy Dimensions of Private vs.
State Enterprise’, The Manchester School, Vol.70, No.2, pp.246–61.
Cook, R., 1977, ‘Detection of Influential Observations in Linear Regression’, Technometrics,
Vol.19, pp.15–18.
Cooley, T. and S. Leroy, 1981, ‘Identification and Estimation of Money Demand’, American
Economic Review, Vol.71, No.5, pp.825–44.
Davis, J., Ossowski, R., Richardson, T. and S. Barnett, 2000, ‘Fiscal and Macroeconomic Impact
of Privatisation’, Occasional Paper, No.194, Washington, DC: IMF.
DeLong, J. and L. Summers, 1991, ‘Equipment Investment and Economic Growth’, Quarterly
Journal of Economics, Vol.106, No.2, pp.445–502.
Deravi, K., Hegji, C. and D. Moberly, 1990, ‘Government Debt and the Demand for Money: An
Extreme Bound Analysis’, Economic Inquiry, Vol.28, No.2, pp.390–401.
Donald, S. and G. Maddala, 1993, ‘Identifying Outliers and Influential Observations in
Econometric Models’, in G. Maddala, C. Rao and H. Vinod (eds.), Handbook of Statistics,
Vol.2, Amsterdam: Elsevier.
Doppelhofer, G., Miller, R. and X. Sala-i-Martin, 2000, ‘Determinants of Long-Term Growth: A
Bayesian Averaging of Classical Estimates (BACE) Approach’, OECD Working Paper, No.266.
Easterly, W. and S. Rebelo, 1993, ‘Fiscal Policy and Economic Growth: An Empirical
Investigation’, Journal of Monetary Economics, Vol.32, No.2, pp.417–58.
Fiebig, D., 1992, ‘Diagnostic Checking in Practice: The Case of Outliers and Influential Data’,
mimeo, University of Sydney.
Folster, S. and M. Henrekson, 2001, ‘Growth Effects of Government Expenditure and Taxation
in Rich Countries’, European Economic Review, Vol.45, No.8, pp.1501–20.
Galal, A., Jones, L., Tandon, P. and I. Vogelsang, 1994, Welfare Consequences of Selling Public
Enterprises: An Empirical Analysis, New York: Oxford University Press.
Gentleman, J. and M. Wilk, 1975, ‘Detecting Outliers II: Supplementing the Direct Analysis of
Residuals’, Biometrics, Vol.31, No.2, pp.387–410.
Harberger, A., 1987, ‘Comment’, NBER Macroeconomics Annual, pp.255–8.
Jensen, M. and W. Meckling, 1976, ‘Theory of the Firm: Managerial Behavior, Agency Costs,
and Ownership Structure’, Journal of Financial Economics, Vol.3, No.4, pp.305–60.
396jds05.qxd 29/08/03 14:13 Page 146

Kalemili-Ozcan, S., Ryder, H. and D. Weil, 2000, ‘Mortality Decline, Human Capital Investment,
and Economic Growth’, Journal of Development Economics, Vol.62, No.1, pp.1–23.
Kikeri, S., 1998, ‘Privatisation and Labor: What Happens to Countries When Governments
Divest’, World Bank Technical Paper, No.396, Washington, DC: World Bank.
Kikeri, S., Nellis, J. and M. Shirley, 1994, Privatisation: The Lessons of Experience, Washington,
DC: World Bank.
Laffont, J. and J. Tirole, 1991, ‘Privatisation and Incentives’, Journal of Law, Economics and
Organisation, Vol.7, No.1, pp.84–105.
Leamer, E., 1978, Specification Searches: Ad Hoc Inference from Non-Experimental Data, New
York: Wiley.
Leamer, E., 1983, ‘Let’s Take the Con Out of Econometrics’, American Economic Review,
Vol.73, No.1, pp.31–43.
Leamer, E., 1985, ‘Sensitivity Analysis Would Help’, American Economic Review, Vol.75, No.3,
pp.308–13.
Leamer, E. and L. Herman, 1983, ‘Reporting the Fragility of Regression Estimates’, Review of
Economics and Statistics, Vol.65, No.2, pp.306–17.
Levine, R. and D. Renelt, 1992, ‘A Sensitivity Analysis of Cross-Country Growth Regressions’,
American Economic Review, Vol.82, No.4, pp.942–63.
Levine, R. and S. Zervos, 1993, ‘What We Have Learned About Policy and Growth from Cross-
Country Regressions?’ American Economic Review, Vol.83, No.2, pp.426–30.
Londregan, J. and K. Poole, 1990, ‘Poverty, the Coup Trap, and the Seizure of Executive Power’,
World Politics, Vol.423, No.2, pp.151–83.
Lorgelly, P. and P. Owen, 1999, ‘The Effect of Female and Male Schooling on Economic Growth
in the Barro-Lee Model’, Empirical Economics, Vol.24, No.3, pp.537–57.
Luke, P. and M. Schaffer, 2000, ‘Wage Discrimination in Russia: An Econometric Investigation’,
William Davidson Institute Working Paper Series, No.295, University of Michigan Business
School.
Mankiw, N., Romer, D. and D. Weil, 1992, ‘A Contribution to the Empirics of Economic
Growth’, Quarterly Journal of Economics, Vol.107, No.2, pp.407–37.
Martin, S. and D. Parker, 1997, The Impact of Privatisation: Ownership and Corporate
Performance in the UK, London: Routledge.
Megginson, W., Nash, R., and M. Van Randenborgh, 1994, ‘The Financial and Operating
Performance of Newly Privatised Firms: An International Empirical Analysis’, Journal of
Finance, Vol.49, No.2, pp.403–52.
Megginson, W and J. Netter, 2001, ‘From State to Market: A Survey of Empirical Studies on
Privatisation’, Journal of Economic Literature, Vol.39, No.2, pp.321–89.
Parker, D., 1999, ‘Privatisation in the European Union: A Critical Assessment of its
Development, Rationale and Consequences’, Economic and Industrial Democracy, Vol.20,
pp.9–38.
Plane, P., 1997, ‘Privatisation and Economic Growth: An Empirical Investigation from a Sample
of Developing Market Economies’, Applied Economics, Vol.29, No.2, pp.161–78.
Ram, R and T. Schulz, 1979, ‘Life Span, Health, Savings, and Productivity’, Economics
Development and Cultural Change, Vol.27, No.3, pp.399–421.
Rousseeuw, P. and A. Leroy, 1987, Robust Regression and Outlier Detection, New York: John
Wiley.
Sala-i-Martin, X., 1994, ‘Cross-Sectional Regressions and the Empirics of Economic Growth’,
European Economic Review, Vol.38, No.3/4, pp.739–47.
Sala-i-Martin, X., 1997, ‘I Just Run Two Million Regressions’, American Economic Review,
Vol.87, No.2, pp.176–83.
Stokey, N., 1994, ‘Comments on Barro and Lee’, Carnegie-Rochester Conference Series on
Public Policy, Vol.40, pp.47–57.
Temple, J., 1999, ‘The New Growth Evidence’, Journal of Economic Literature, Vol.XXXVII,
pp.112–56.
Tullock, G., 1965, The Politics of Bureaucracy, Washington, DC: Public Affairs Press.
UNCTAD, 1996, Comparative Experiences with Privatisation: Policy Insights and Lessons
Learned, New York: United Nations.
396jds05.qxd 29/08/03 14:13 Page 147
Vickers, J. and G. Yarrow, 1988, Privatisation: An Economic Analysis, London: MIT Press
Yarrow, G., 1999, ‘Theory of Privatisation, or Why Bureaucrats are still in Business’, World
Development, Vol.27, No.1, pp.157–68.
Welsch, R., 1982, ‘Influence Functions and Regression Diagnostics’, in R. Launer and A. Siegel
(eds.), Modern Data Analysis, New York: Academic Press, pp.149–69.
Welsch, R. and E. Kuh, 1977, ‘Technical Report’, in Linear Regression Diagnostics, 923-77,
Cambridge, MA: Sloan School of Management, MIT.
World Bank, 1994, World Bank Assistance to Privatization in Developing Countries, OED
Report, No.3273, Aug.
World Bank, 1995a, Bureaucrats in Business: The Economics and Politics of Reform, Oxford:
Oxford University Press, for World Bank.
World Bank, 1995b, Structural and Sectoral Adjustment Lending: World Bank Experience
(1980–1992), Operations Evaluation Department, Washington, DC: World Bank.
APPENDIX 1
Consider the following three variable linear regression:

yt = xtβ + z1tγ1 + z2tγ2 + ut (1)
where β is the coefficient of particular interest to be estimated, x is the variable of particular
interest or the variable that is always included in a regression, z denotes the ‘doubtful’ variables,
and γ is the corresponding coefficient to be estimated. Here, the z variables represent specification
uncertainty, and cannot be excluded a priori.
A Bayesian approach assumes that it is possible to define a prior distribution with the prior
location of such doubtful variables and to specify their covariance matrix, thereby obtaining a
posterior distribution for β. In practice, however, it is difficult to specify such a matrix. Leamer
(1978) has developed a theorem in which specification of the prior covariance matrix is not
required, assuming that specification of the prior location and the sample covariance matrix is
only necessary to constrain the posterior means to lie within a certain ellipsoid or the locus of
constrained estimates.
Suppose that the object is to show that β is large or that the estimation of γ1 and γ2 is difficult.
In a conventional approach, four different regressions would be computed using different
combinations of the control variables, z1 and z2. Then, among the four different results obtained,
the most favourable result would be reported. What Leamer has developed is an alternative
procedure in which the specification search is enlarged and reports consist of the most favourable
and least favourable results.
To enlarge the search for more specifications, a composite variable is defined as follows:
wt(θ) = z1t + θz2t, (2)
where θ is a number to be selected and is allowed to take any value.
Then, equation (1) can be rewritten as:
yt = xtβ + wt(θ)γ + ut. (3)
The search can be enlarged by including all values of θ. For each value of θ, a different
constraint of the form, γ2 = θγ1 is imposed, and there exits a least squares estimate of β, β(θ). By
maximising β(θ) with respect to θ, the most favourable value of θ can be obtained, and the least
favourable value can be obtained by minimising the value.
Figure A1 shows the theorem graphically. In a conventional approach, four different
regressions are estimated. These are indicated by the four points, (γ1*,γ2*), (γ1*,0), (0,γ2*), (0,0)
in the figure. However, there is no reason why the search should be restricted by only estimating
these four regressions. To enlarge the search, any values for θ can be used and a different
regression is estimated and the corresponding focus coefficient β* (θ) for each value of θ. The
dished line in the figure depicts the ellipse of constrained estimates defined by the set of all
possible values of (γ1,γ2) that are obtained by changing the value of over the real line. Thus, the
ellipse contains all posterior means for the distribution of (γ1, γ2 ).
396jds05.qxd 29/08/03 14:13 Page 148
However, the extreme values, max β* (θ) and min β* (θ), may be located in an unlikely
parameter space on the locus of constrained estimates in relation to the data. To resolve this
problem, define a statistically feasible likelihood ellipse, say at 0.05 level or 95 per cent, which
is depicted in the dashed lines in the figure, within the likelihood ellipse. This is defined by the
assumption that all doubtful variables are included in the regression. The intersection of the
points in the interior of the ellipse of constrained estimates and this likelihood ellipse is the set of
points (γ1*, γ2*). The area defined by the set is shown by the shaded area in the figure.
Thus, the set of all posterior means for the distribution of (γ1*, γ2*) is here subject to further
constraint in this area, eliminating the unlikely parameter space. The parameter pairs in the area
can be obtained by posterior estimates from prior distribution centred at the origin, and the
measure of specification uncertainty is the difference between the extreme values of β* (θ) over
the area.
FIGURE A1
ELLIPSOIDS OF CONSTRAINED LEAST SQUARES POINTS
APPENDIX 2
PRIV GYP LGYP LIFE POP CON STAB OPEN GDI BUD FDI IMP SDIMP INFL SDINF M3 SDM3 DEBT
Argentina 0.10 0.02 8.58 71.01 0.01 0.04 6.00 0.16 0.17 –0.01 0.01 0.22 0.24 5.65 10.78 0.18 0.05 0.04
Bangladesh 0.00 0.03 7.17 53.30 0.02 0.04 1.00 0.23 0.19 –0.06 0.00 0.80 0.46 0.05 0.03 0.26 0.03 0.02
29/08/03
Barbados 0.03 0.00 8.91 74.34 0.00 0.20 0.00 0.98 0.15 –0.02 0.01 0.08 0.07 0.02 0.02 0.62 0.06 0.07
Belize 0.07 0.03 8.09 72.57 0.03 0.15 0.00 1.16 0.26 –0.03 0.03 0.15 0.10 0.03 0.02 0.43 0.03 0.05
Benin 0.01 0.01 6.90 50.85 0.03 0.11 1.00 0.58 0.16 –0.07 0.00 0.03 0.01 0.07 0.11 0.26 0.03 0.02
Bolivia 0.05 0.02 7.42 56.81 0.02 0.13 3.00 0.48 0.15 –0.03 0.03 0.02 0.02 0.12 0.04 0.35 0.11 0.06
14:13
Brazil 0.04 0.01 8.34 64.76 0.02 0.17 4.00 0.17 0.21 –0.08 0.01 0.29 0.32 10.21 9.53 0.44 0.18 0.03
Burkina Faso 0.00 0.01 6.26 45.87 0.02 0.14 6.00 0.38 0.22 0.00 0.00 0.03 0.01 0.05 0.09 0.23 0.01 0.02
Burundi 0.00 –0.03 6.33 46.71 0.02 0.12 6.00 0.35 0.13 –0.06 0.00 0.27 0.13 0.10 0.07 0.19 0.02 0.04
Cameroon 0.00 –0.04 7.19 53.31 0.03 0.11 2.00 0.40 0.16 –0.03 0.00 0.03 0.01 0.04 0.06 0.19 0.04 0.05
Chile 0.01 0.06 8.29 72.67 0.02 0.10 0.00 0.61 0.25 0.03 0.04 0.13 0.06 0.13 0.07 0.41 0.02 0.08
Colombia 0.03 0.02 8.08 68.23 0.02 0.13 3.00 0.34 0.20 –0.01 0.02 0.09 0.05 0.24 0.03 0.32 0.07 0.08
Page 149
Costa Rica 0.00 0.02 8.12 74.84 0.02 0.16 0.00 0.82 0.26 –0.01 0.03 0.02 0.07 0.18 0.05 0.41 0.03 0.08
Cote d’Ivoire 0.02 –0.01 7.26 50.96 0.03 0.15 1.00 0.68 0.11 –0.14 0.01 0.03 0.01 0.05 0.13 0.28 0.01 0.13
Ecuador 0.00 0.01 7.95 67.04 0.02 0.10 2.00 0.57 0.20 –0.02 0.02 0.13 0.10 0.42 0.16 0.27 0.06 0.09
Egypt, Arab Rep. 0.02 0.02 7.55 61.07 0.02 0.11 2.00 0.53 0.22 –0.04 0.02 0.07 0.19 0.13 0.05 0.87 0.03 0.06
Ghana 0.04 0.02 6.70 55.96 0.03 0.11 0.00 0.50 0.18 –0.04 0.01 0.09 0.10 0.29 0.10 0.17 0.02 0.07
Guatemala 0.00 0.01 7.66 59.69 0.03 0.06 12.00 0.42 0.15 –0.01 0.01 0.10 0.09 0.16 0.11 0.25 0.02 0.03
Guinea 0.00 0.02 6.65 42.55 0.03 0.10 0.00 0.49 0.19 –0.07 0.00 0.07 0.05 0.12 0.11 0.07 0.04 0.04
Guinea-Bissau 0.00 0.02 6.48 41.56 0.02 0.08 1.00 0.51 0.31 –0.26 0.00 0.45 0.20 0.54 0.24 0.16 0.02 0.05
Honduras 0.01 0.01 7.26 65.40 0.03 0.13 3.00 0.75 0.28 –0.06 0.02 0.26 0.30 0.18 0.08 0.33 0.03 0.11
India 0.01 0.04 7.09 57.85 0.02 0.11 11.00 0.21 0.23 –0.06 0.00 0.10 0.04 0.09 0.03 0.48 0.02 0.03
Indonesia 0.01 0.06 7.46 60.21 0.02 0.08 1.00 0.51 0.30 0.00 0.01 0.07 0.08 0.09 0.02 0.43 0.08 0.09
Iran, Islamic Rep. 0.00 0.02 8.08 64.73 0.02 0.12 0.00 0.39 0.28 –0.02 0.00 2.29 1.19 0.27 0.09 0.53 0.09 0.04
Jamaica 0.07 0.01 7.80 72.41 0.01 0.14 2.00 1.19 0.31 –0.03 0.03 0.18 0.09 0.28 0.16 0.53 0.03 0.18
Jordan 0.00 –0.02 8.16 66.71 0.04 0.25 0.00 1.32 0.31 –0.02 0.00 0.05 0.03 0.06 0.06 1.18 0.15 0.13
Kenya 0.01 0.00 6.80 57.55 0.03 0.16 3.00 0.62 0.21 –0.05 0.00 0.14 0.16 0.13 0.07 0.47 0.05 0.10
Korea, South 0.01 0.07 8.63 69.27 0.01 0.10 0.00 0.64 0.36 –0.01 0.00 0.01 0.05 0.06 0.02 0.67 0.14 0.03
Malawi 0.00 0.01 6.20 44.88 0.03 0.16 0.00 0.60 0.19 –0.07 0.00 0.29 0.16 0.29 0.24 0.23 0.04 0.06
Malaysia 0.08 0.06 8.36 69.50 0.03 0.13 1.00 1.65 0.36 –0.03 0.06 0.01 0.01 0.04 0.01 1.05 0.27 0.10
A P P E N D I X 2 (cont.)
Mali 0.00 0.01 6.25 46.51 0.03 0.13 2.00 0.54 0.20 –0.05 0.00 0.03 0.01 0.06 0.10 0.22 0.01 0.03
Mauritania 0.00 0.01 6.67 49.46 0.03 0.13 0.00 1.02 0.20 –0.02 0.01 0.70 0.57 0.06 0.03 0.23 0.05 0.12
29/08/03
Mexico 0.05 0.01 8.58 69.82 0.02 0.10 0.00 0.44 0.23 0.00 0.02 0.05 0.05 0.31 0.29 0.25 0.07 0.08
Morocco 0.02 0.01 7.66 62.01 0.02 0.17 2.00 0.57 0.22 –0.04 0.01 0.05 0.04 0.04 0.02 0.66 0.08 0.10
Mozambique 0.01 0.03 6.63 43.47 0.02 0.13 6.00 0.52 0.17 –0.27 0.01 0.25 0.20 0.43 0.14 0.33 0.05 0.06
Nepal 0.00 0.03 6.91 52.02 0.03 0.09 1.00 0.46 0.22 –0.08 0.00 0.30 0.11 0.10 0.04 0.35 0.03 0.02
14:13
Nicaragua 0.02 –0.01 7.27 62.17 0.03 0.22 7.00 0.80 0.23 –0.15 0.02 0.41 0.63 31.11 45.44 0.36 0.16 0.13
Nigeria 0.01 0.02 6.85 48.31 0.03 0.13 3.00 0.74 0.18 –0.10 0.04 0.64 0.50 0.36 0.24 0.20 0.05 0.09
Oman 0.00 0.01 8.89 67.96 0.04 0.33 0.00 0.86 0.16 –0.09 0.01 0.02 0.01 0.00 0.09 0.31 0.02 0.07
Pakistan 0.01 0.02 7.22 58.24 0.03 0.14 6.00 0.36 0.16 –0.08 0.01 0.06 0.03 0.10 0.02 0.44 0.03 0.05
Panama 0.06 0.01 7.94 71.74 0.02 0.17 11.00 0.73 0.19 0.02 0.02 0.00 0.00 0.02 0.02 0.52 0.13 0.07
Papua New Guinea 0.05 0.01 7.39 53.93 0.02 0.20 9.00 0.97 0.25 –0.03 0.04 0.10 0.06 0.05 0.06 0.34 0.03 0.14
Page 150
Paraguay 0.00 0.01 7.61 67.60 0.03 0.08 0.00 0.82 0.23 0.01 0.02 0.23 0.25 0.20 0.09 0.26 0.05 0.05
Peru 0.09 0.00 7.91 64.39 0.02 0.08 12.00 0.26 0.21 –0.02 0.02 0.36 0.49 10.13 20.11 0.19 0.04 0.03
Philippines 0.02 0.01 7.42 64.00 0.02 0.11 16.00 0.72 0.23 –0.04 0.02 0.05 0.02 0.09 0.03 0.46 0.12 0.07
Senegal 0.02 0.00 7.07 48.25 0.03 0.14 0.00 0.62 0.15 –0.01 0.01 0.03 0.01 0.04 0.24 0.23 0.01 0.06
Sierra Leone 0.00 –0.05 6.79 36.81 0.02 0.10 1.00 0.41 0.07 –0.06 0.00 0.88 0.93 0.48 0.27 0.13 0.03 0.06
Singapore 0.08 0.07 9.24 73.59 0.02 0.09 0.00 3.68 0.35 0.12 0.10 0.02 0.01 0.04 0.02 1.18 0.04 0.11
South Africa 0.02 –0.01 8.12 60.43 0.02 0.20 1.00 0.48 0.17 –0.05 0.01 0.04 0.02 0.12 0.03 0.46 0.02 0.03
Sri Lanka 0.02 0.04 7.61 70.70 0.01 0.10 12.00 0.73 0.24 –0.10 0.01 0.13 0.10 0.11 0.03 0.40 0.02 0.04
Tanzania 0.01 0.00 6.28 51.01 0.03 0.17 0.00 0.51 0.22 –0.09 0.01 0.30 0.23 0.26 0.08 0.24 0.03 0.04
Thailand 0.00 0.07 8.00 67.49 0.01 0.10 5.00 0.80 0.39 0.02 0.02 0.03 0.01 0.05 0.01 0.79 0.07 0.06
Togo 0.01 0.00 6.46 51.41 0.03 0.13 8.00 0.75 0.16 –0.05 0.00 0.03 0.01 0.07 0.11 0.32 0.05 0.04
Trinidad and 0.06 0.00 9.00 70.39 0.01 0.13 2.00 0.83 0.16 –0.02 0.05 0.23 0.18 0.07 0.07 0.53 0.05 0.10
Tobago
Tunisia 0.00 0.02 7.90 65.63 0.02 0.16 0.00 0.89 0.26 –0.04 0.02 0.05 0.05 0.06 0.02 0.50 0.02 0.10
Turkey 0.01 0.02 8.14 64.25 0.02 0.11 5.00 0.39 0.24 –0.05 0.00 0.03 0.04 0.75 0.15 0.29 0.04 0.06
Uganda 0.01 0.04 6.27 48.34 0.03 0.09 2.00 0.30 0.14 –0.07 0.01 0.38 0.40 0.48 0.60 0.10 0.01 0.04
Uruguay 0.00 0.02 8.45 72.11 0.01 0.13 2.00 0.43 0.13 –0.03 0.00 0.12 0.08 0.58 0.28 0.49 0.09 0.06
Venezuela 0.04 0.00 8.82 70.53 0.02 0.08 8.00 0.54 0.18 –0.02 0.02 0.19 0.36 0.50 0.31 0.35 0.08 0.08
Vietnam 0.00 0.05 6.79 65.80 0.02 0.08 0.00 0.69 0.21 0.00 0.04 0.46 0.73 0.70 1.23 0.22 0.01 0.03
Yemen 0.00 0.00 7.54 51.49 0.04 0.19 0.00 0.61 0.20 –0.08 0.04 0.17 0.07 0.25 0.16 0.51 0.07 0.51
29/08/03
Zambia 0.05 –0.01 6.63 49.63 0.03 0.17 1.00 0.72 0.13 –0.11 0.02 0.69 0.68 0.77 0.50 0.22 0.07 0.17
Zimbabwe 0.01 0.01 6.99 56.91 0.03 0.19 0.00 0.62 0.21 –0.09 0.00 0.29 0.16 0.21 0.06 0.40 0.05 0.08
Mean 0.02 0.02 7.51 59.83 0.02 0.13 3.06 0.67 0.21 –0.05 0.02 0.22 0.18 1.09 1.48 0.39 0.06 0.08
14:13
Standard Deviation 0.03 0.02 0.80 9.95 0.01 0.05 3.88 0.47 0.06 0.06 0.02 0.33 0.24 4.28 6.40 0.23 0.05 0.07
Notes: PRIV = the magnitude of privatisation as the ratio of cumulative proceeds from privatisation during the sample period 1988-97 to average GDP
during the same period; primary data for the proceeds are taken from World Bank Privatization Database and Privatisation International
(various issues); and the data for GDP are taken from World Bank Global Development Network Growth Database.
GYP = the average real GDP per capita (in constant US dollars, international prices, base year 1985) growth rate during the sample period; and
Page 151
the data are taken from World Bank Global Development Network Growth Database.
LGYP = the log of initial real GDP per capita in 1988 capita (in constant US dollars, international prices, base year 1985); and primary data are
taken from World Bank Global Development Network Growth Database
POP = the average population growth rate during the sample period; and the data are taken from World Bank Global Development Network
Growth Database
LIFE = the initial life expectancy at birth in 1987 (the data in 1988 are not available); and the primary data are taken from World Bank Global
Development Network Growth Database
STAB = political instability calculated as the average number of major political crises, revolutions, and coups during the sample period; and the
primary data are taken from World Bank Global Development Network Growth Database
BUD = the average ratio of central government budget surplus/deficit to GDP (calculated in national currency, constant price) during the sample
period; and the primary data are taken from World Bank Global Development Network Growth Database
OPEN = openness of a country calculated as (export-import)/GDP; and the primary data are taken from World Bank Global Development Network
Growth Database
CON = the average ratio of government consumption to GDP (calculated in national currency, constant price) during the sample period; and the
primary data are taken from World Bank World Development Indicators
GDI = the average ratio of gross domestic investment to GDP (calculated in national currency, constant price) during the sample period; and the
primary data are taken from World Bank World Global Development Network Growth Database
152
FDI = the average ratio of foreign direct investment (net inflows) to GDP (calculated in current US dollars) during the sample period; and the
primary data are taken from World Bank World Development Indicators
INFL = the average inflation rates (GDP deflator) during the sample period; and the primary data are taken from World Bank World Development
Indicators
29/08/03
SDINF = the standard deviation of INFL

IMP = log(1+ informal market premium); informal market premium is calculated using the following formula: (Parallel Exchange Rate/Official
Exchange Rate -1)*100; the primary data are taken from World Bank Global Development Network Growth Database
14:13
SDIMP = the standard deviation of IMP

M3 = the ratio of liquid liabilities (M3) to GDP, the primary data are taken from World Bank Global Development Network Growth Database
SDM3 = the standard deviation of M3
DEBT = the ratio of total external debt to GDP, the primary data are taken from World Bank Global Development Network Growth Database.
Page 152
* World Bank Global Development Network Growth Database is available from www.worldbank.org/research/growth/GDNdata.htm
THE J OURNA L O F D E V E L O P ME N T S T U D I E S
396jds05.qxd 29/08/03 14:14 Page 153
APPENDIX 3
Influential Observations/Outlier Detection Techniques:

In general outliers are categorised into two types. The first is characterised as ∧
the
substantially large residual arising from the difference between the actual Yi and the fitted Yi. The
second is that Xi is far from the cluster of Xi’s, resulting in the direction of the fitted line varying,
although this is not indicated by a high residual. Detecting more than one outlier in a dataset is,
in practice, difficult owing to the so-called masking and swamping effects [Luke and Wilk, 2000].
The masking effect occurs when a subset of outliers is not identified due to the presence of
another subset of observations. The swamping effect occurs when normal observations are
misidentified as outliers due to the presence of another subset of observations.
One way to detect influential observations/outliers is to analyse the size of residuals from the
regression estimation. This can be done by computing studentised residuals. The studentised
residual can be computed as:
SR = e / (s
i i (i)
1 − hi ),
where hi is the ith diagonal element of the hat matrix H, s is the variance from sample residuals,
and ei is the ordinary residual. (i) is the standard error obtained by excluding the ith observation,
and then s(i) is calculated as:
(n − k ) s 2 − ei2 /(1 − hi ) 2  n − k − eˆi2 
s(i)2 = s  ,

n − k −1  n − k −1 
where k is the number of explanatory variables and n is the number of observations in the
regression.
DFITS, Cook’s Distance, and Welsch Distance summarise the information in the leverage
versus residual-squared plot into a single statistics by creating an index that is influenced by the
size of the residuals or outliers and the size of leverage.
DEFITS can be calculated by:
DFITSi = hi ,
ri
1 − hi
where ri are the studentised residuals and hi are the size of leverage.
Cook’s Distance is calculated by:
CDi = 1/k • si2/s2 • DFITSi2 ,
where s is the root mean square error of the regression, and si is the root mean square error when
the ith observation is eliminated.
Welsch Distance can be estimated by:
WDi = DFITSi n −1 ,
1 − hi
Then, outliers or the observations that ought to be examined further are defined as:
DFITS > 2 , k n , Cook’s Distance > 4/n, and Welsch Distance > 3 k .
396jds05.qxd 29/08/03 14:14 Page 154

∧ ∧
DFBETAs measure the effect on each coefficient by testing β j – β (i)j , which is defined as:
βˆ j − βˆ (i ) j ,
DFBETAsij =
s ( i ) ( X ' X ) −jj1
where is the (X′ X)–1jj is the jth diagonal element of (X′ X)–1. The observations that ought to be
examined further are defined as DFBETAs > 2 / n .

Privatization and Economic Growth

Загружено:

Сведения о документе

Исходное описание:

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Privatization and Economic Growth

Загружено:

Авторское право:

Доступные форматы

This article was downloaded by:[Hiroshima University]

On: 25 July 2008

Journal of Development Studies

Online Publication Date: 01 August 2003

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf

Privatisation and Economic Growth in

This article re-examines the relation between privatisation and

Privatisation accelerated in the 1990s. In Europe this was sparked by the

122 THE J OURNA L O F D E V E L O P ME N T S T U D I E S

without much knowledge of either its impact or contribution to economic

Policy-makers in developing economies have often set a broader agenda for

public ownership led to the pursuit of objectives that detracted from

124 THE J OURNA L O F D E V E L O P ME N T S T U D I E S

In most developing countries privatisation has been a new policy instrument

long period, to permit the coefficients on various variables to be interpreted

126 THE J OURNA L O F D E V E L O P ME N T S T U D I E S

Several potential problems relating to EBA have been identified [Sala-

Country CPP AGDP CPP/ Country CPP AGDP CPP/

1 Argentina 27921.0 28267.3 9.9 33 Tanzania 140.8 15031.3 0.9

5 Belize 54.3 793.0 6.8 37 Kenya 235.1 28329.6 0.8

128 THE J OURNA L O F D E V E L O P ME N T S T U D I E S

privatisation variable may be linked to the size of the government budget

received adjustment loans with explicit conditions to implement

130 THE J OURNA L O F D E V E L O P ME N T S T U D I E S

Of course, these results do not automatically confirm that the

IV. EXTREME BOUNDS ANALYSIS (EBA )

The Basic Test for the Control Variables (the I variables)

investment is often used as a proxy for investment. However, Barro and

132 THE J OURNA L O F D E V E L O P ME N T S T U D I E S

the average ratio of government budget surplus/deficit to GDP (BUD); and

Variable Coefficient t-statistic Z-variables included EBA result

LGYP Min –0.012 –2.05 STAB, EA, AF Fragile

Variable Coefficient t-statistic Z-variables included EBA result

LGYP Min – – – Insignificant

Variable Coefficient t-statistic Z-variables included EBA result

LGYP Min –0.013 –2.34 STAB, EA, AF Fragile

LIFE Min 0.052 2.01 GDI, SDIMP, LA Robust

134 THE J OURNA L O F D E V E L O P ME N T S T U D I E S

Specification III (cont.)

Variable Coefficient t-statistic Z-variables included EBA result

GOVC Min –0.203 –3.17 STAB, BUD, M3 Robust

Variable Coefficient t-statistic Z-variables included EBA result

LGYP Min –0.015 –2.69 STAB, EA, AF Fragile

POP Min –1.216 –3.46 FDI, SDM3, LA Robust

The EBA for Privatisation

examined previously and the privatisation variable. The Z variables were

The I variables based on Specification I (LGYP + LIFE + POP + GDI)

Variable Coefficient t-statistic Z-variables included EBA result

PRIV Min –0.243 –2.25 BUD, FDI, DEBT Fragile

The I variables based on Specification II (LGYP + LIFE + POP + GDI + GOVC)

Variable Coefficient t-statistic Z-variables included EBA result

PRIV Min –0.249 –2.35 BUD, FDI, DEBT Fragile

The I variables based on Specification III (LGYP + LIFE + POP + GOVC)

Variable Coefficient t-statistic Z-variables included EBA result

PRIV Min –0.347 –2.82 FDI, SDM3, DEBT Fragile

The I variables based on Specification IV (LGYP + LIFE + POP)

Variable Coefficient t-statistic Z-variables included EBA result

PRIV Min –0.345 –2.76 FDI, SDM3, DEBT Fragile

136 THE J OURNA L O F D E V E L O P ME N T S T U D I E S

138 THE J OURNA L O F D E V E L O P ME N T S T U D I E S

Studentized DFBETAs DFITS Cook’s Welsch

Studentized DFBETAs DFITS Cook’s Welsch

PRIV –0.090 –0.192* –0.254* –0.292* –0.063 –0.178 –0.214** –0.223*

LIFE 0.069 0.068* 0.032 0.036 0.085* 0.089* 0.051* 0.085***

PRIV –0.067 –0.106 –0.250* –0.285* –0.094 –0.204* –0.221* –0.249**

PRIV –0.067 –0.271 –0.288 –0.249**