Privatization and economic growth in the 1990s A panel data approach

(1)

Privatization and economic growth in the

1990s

A panel data approach

J.J. Rodenburg

Rijksuniversiteit Groningen

Master Thesis

(2)

Privatization and economic growth in the

1990s

A panel data approach

J.J. Rodenburg

Rijksuniversiteit Groningen

Abstract

(3)

Preface

(4)

1. Introduction

“There Is No Alternative”1

This famous quote by Margaret Thatcher, by which she argued that free markets, free trade and capitalist globalization are the only direction for modern societies, initiated a privatization wave in the United Kingdom (UK). Although the Thatcher government was not the first to adopt a large privatization program, its perceived success helped encourage many other governments worldwide in launching similar programs. Before the UK privatization program, depressions and wars raised the belief that governments should play an active role in regulating the national economy. The level of involvement differed among governments but the general belief was that the government should at least play a significant role in sectors like social security, wage setting, telecommunications, electricity and gas, and transport. After the UK program, many other countries started to embrace privatization. The first was France and not much later other European countries, quickly followed by countries in Asia. The Washington Consensus led to a privatization wave in Latin America in the early nineties. In South East Asian and African countries, followed by countries in East Europe, the last major region to adopted privatization programs, privatization has been a part World Bank and International Monetary Fund (IMF) conditionality for receiving financial aid.

(7)

2005). In addition Kikeri and Kolo (2005) find that recent studies have shown that privatization in for instance social sectors also leads to efficiency gains. In this research the gains of both regular companies and utilities are combined in one sample and are treated identically.

Ever since the privatization wave started extensive research has been done on the effects of it. A lot of empirical literature addresses to the essentially microeconomic question whether privatization increased performance, expressed in productivity, efficiency, service and employment. The foundation of this research area is built by Boardman and Vining (1989) and Megginson et al. (1994). The overall conclusion is that privatization has a positive effect on these variables and thus country performance2. Intuitively speaking and simply put, when single companies profit from privatization so should a country as a whole. This due to the cumulative effects i.e. privatization should enhance economic growth. However it has not been easy to prove this relation empirically. The research that has been conducted on the macroeconomic effects of privatization is mainly aimed at the fiscal effects of privatization. This is mainly because of the fact that fiscal effects do occur immediately and therefore this direct effect can be investigated relatively easy. To investigate the long run effects of privatization on growth has proven to be more complicated since a macroeconomic variable like economic growth is influenced by so many factors that it is hard to filter out the effects of privatization. In addition these factors often have influences on each other. Another issue is the puzzle of privatization itself. If privatization is such a panacea, what are the arguments of the opponents? Or even, stronger why are there opponents? Common arguments against privatization are that it leads to unemployment and negatively influences the income distribution in a country.

This research investigates the effects of privatization on long term economic growth. First of all to see if there is any effect and whether this effect is positive or negative. Secondly, to see whether the effects differ with the level of economic development and the kind of economic system.

This study differentiates itself from others because it makes us of a panel based data analysis. Panel analysis makes it possible to directly link privatization to economic growth in a set of countries and at the same time analyze this relation for a time series. Other studies that investigated the effects of privatization on economic growth made use of cross-country regression analysis averaging variables over periods of time (Plane 1997, Cook and Uchida 2001). This due to the fact that in the samples they used no yearly data were available.

2

(8)

The large number of countries that are included in this research increases the number of privatization data points to a level that, when pooled together, a time-series analysis becomes possible.

To give an overview of the size of privatization figure 1.1 gives the total proceeds of privatization in developing countries provides by the World Bank Privatization Database and European countries by the Privatization Barometer for the years 1990 to 2003. As can be seen privatization flourished at the end of the nineties.

Figure 1.1 Total privatization proceeds in current US dollars

0 10,000 20,000 30,000 40,000 50,000 60,000 70,000 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 Year U S $ m il li o n s

Total (all regions)

Source: World Bank Privatization Database & Privatization Barometer

(9)

Figure 1.2 Total cumulative proceeds top ten (current US dollars) Sum of value of transactions us $ millions

0 20,000 40,000 60,000 80,000 100,000 120,000 140,000 160,000 180,000 Italy France United Kingdom Germany Brasil Spain Argentina China Mexico C o u n tr y US $ millions

Source: World Bank Privatization Database & Privatization Barometer

While figure 1.2 gives an overview of the vast amounts that government receive this does not say a whole lot about the size of privatization in a country. To give an impression of the scale of privatization, in figure 1.3 privatization proceeds are given as a share of the country’s gross domestic product (GDP). Here we can see that, over years, the Slovak Republic has privatized over 40 percent of its nations GDP, while a country like The Netherlands (not stated in the figure) has privatized for a total of 8 percent of a year’s GDP.

Figure 1.3 Total Proceeds divided by GDP top ten

Total proceeds (1990-2003) / GDP (2000) 0.00 0.10 0.20 0.30 0.40 0.50 Slovak Republic Kazakhstan Bulgaria Hungary Macedonia, FYR Czech Republic Serbia and Montenegro Portugal Zambia Morocco

(10)

1.1 Research objectives

The objective of this research is to investigate the macroeconomic effects of privatization. The corresponding research question is the following: What is the impact of privatization on economic

growth? Is this impact positive or negative? Are the effects stronger among more developed countries or is the reverse true: do poorer countries profit more from privatization because they are relatively less efficient and privatization provides a way to catch up? Could another factor, such as the economic system of a country be of influence? To find, with empirical research and statistical analyses, an answer to these questions, a number of hypotheses are examined in this research. First I examine the relation between privatization and economic growth. A positive relation is expected here. The main argument here is that on firm level privatization has positive productivity, efficiency and employment and thus performance effects as Boardman and Vining (1992) find that privatization from a efficiency perspective is preferable towards state-ownership. All these positive effects together added up with the extra funds governments receive to invest either in reducing debt or investing in new infrastructure should lead to positive growth effects in a country. The first hypothesis therefore is:

Hypothesis 1. Privatization is positively related with economic growth.

Secondly, how does the development of a country influence this relation. Intuitively, developing countries should be able to profit more from privatization than developed countries since they are relatively less efficient than developed countries and therefore privatization can be a means to catch up with the developed countries. Dewenter and Malatesta (2001) find that privatization should induce governments to take great efficiency steps thereby enlarging the value of their companies before selling them. Furthermore, Frydman, Gray, Hessel and Papaczynski (1997) suggest that privatization is most important when the environment of firms is uncertain. This leads to the following hypothesis:

Hypothesis 2. Economic development weakens on the relation between privatization and

economic growth.

(11)

arrangements. SOEs in these economies are expected to function in similar way. From this line of reasoning it follows that SOEs in more liberal economies are relatively more similar to private organizations than SOEs in more coordinated economies. In addition, private organizations in more coordinated market economies remain more restrained by their government. When institutional reforms come with privatization this can open up additional potential for growth. Therefore, more liberal economies should profit relatively less from privatization than more coordinatedcountries as their SOEs.

This leads to the following hypothesis:

Hypothesis 3. The economic system of a country influences the relation between

privatization and growth, where in a more liberal economic system the effect is smaller.

A graphical interpretation of the hypotheses can be found in figure 4.

Figure 4. Visual interpretation of the hypotheses

(12)

(13)

2. Theoretical background

Here an overview of the relevant literature is given to build the theoretical foundation of the hypotheses. First the theory on privatization is discussed, followed by an overview of the important empirical research performed on this subject. Thereafter the theory on economic systems is covered. Since in this research the relation of privatization on growth is investigated, a summary of the relevant theory on economic growth as well as an overview of the empirical research on growth explaining variables is given.

2.1 Privatization theory

To state a definition for privatization I would like to refer to the one given by Kwoka (1996). He states that the term privatization refers to the shift from public to private provision of goods and services. This can be done by transfer of ownership of an enterprise to private investors through asset sale, equity sale or equity distribution. It does not matter whether this is to a single owner, a group of investors, the public at large or even the enterprise’s employees.

In its essence privatization is a response to the failures of state ownership or the result of political motives. Under the conditions of perfect competition, nonexistence of information problems and complete contracts, ownership does not matter, and thus there should be no difference between state and private ownership. In this case there should also be no room for privatization. Theoretically, governments should only intervene in the case of market failure i.e. failure of the first condition mentioned above. Real world existence of externalities, public goods, monopolistic markets and information costs, were, for a long time, a reason to allow a relatively small government role restricted to allowing trade and protecting property rights.

However government intervention is involved with efficiency losses, due to the existence of information asymmetries and incomplete contracting problems, and these turned out to be significant. This provided the basis for privatization. The theory on this subject can be divided into two parts as showed by Sheshinski and López-Calva (2003). On the one hand the problems associated with public ownership, or the agency view and on the other hand the macroeconomic effects of privatization.

2.1.1 Agency view

(14)

to diffuse ownership. In addition, governments do not have objectives like profit or shareholder wealth maximization. It can be argued that for a SOE objectives like profit or shareholder wealth maximization can be harmful (for instance, it could lead to undesirable mergers). Nevertheless it brings along less incentive for efficiency. Even if one argues that the shareholder wealth maximization for a government is maximizing social welfare, these objectives need not be consistent with efficiency and above all are difficult to measure. Secondly, managers in public firms are likely to have different objectives than shareholder maximization. For the shareholders of a country (its citizens) it is hard to check if managers are acting to their benefit, leaving incentives for these managers to act in their personal interest. In addition because they are owned by the state SOEs are unlikely to go bankrupt. According to Megginson and Netter (2001) discipline enforced by capital markets and the threat of financial distress favor private ownership. Zinnes et al. (2001) summarize the above named arguments nicely by stating that the gains of change from public to private ownership depend on the issues of OBCA, where the O stands for the firm’s objective function, BC for the firms budget constraint and A for the so-called principal agent problem.

2.1.2 Macroeconomic effects

Sheshinski and López-Calva (2003) consider several macroeconomic effects. The first are the immediate and long run fiscal effects. Privatization can create large funds in the short run3 and eliminates subsidies to loss-making SOEs in the long run. Actually, privatization can lead to tax increases when these firms become profit making due to restructuring. Furthermore, macro instability accelerates privatization i.e. higher public deficits lead to faster restructuring programs. More importantly privatization often was (and still is) a part of IMF loan conditionality and World Bank assistance. A second effect is the development of the financial sector and more specific the level of stock market capitalization as capital is mobilized to areas where it can be invested with higher returns. It needs to be said that this argument is not very strong, as up till now no empirical evidence for this relation is found (Levine and Zervos, 1998). Finally Zinnes et al. (2001) indicate privatization has opposite effects on employment in the short and long run. In the process of privatization redundant labor is dismissed and unemployment may actually increase in the short run. However in the long run employment is expected to increase as the growth increases through efficiency gains and greater stability. Furthermore Zinnes et al. (2001)

3

(15)

stress that the gains of privatization at the macroeconomic level depend on complementary policies.

The theoretical arguments mentioned above clearly assume a positive relation of privatization on growth, as privatization should be beneficial both at firm and macro level.

But as stated in the introduction, if privatization is If privatization is such a panacea, what are the arguments of the opponents? An argument of opponents of privatization is that the gains of privatization are outweighed by the welfare losses to the public and the economy as a whole. These losses come in the form of decreased access to products and services, suboptimal supply and disproportionate high prices. Kikeri and Nellis (2004) report that there are only a few studies on the broader welfare effects of privatization. This is mainly caused by lack of data. Such studies rely on extensive data, which often is not available. Nevertheless Kikeri and Nellis (2004) report that the studies on the broader welfare effects all find improved economic welfare after privatization. In addition, there are several partial welfare analyses. First of all, considering social effects, Kikeri and Nellis (2004) conclude that almost all studies find that consumer surplus expanded after privatization. Regarding the effect on access Kikeri and Nellis (2004) find that most studies report increased access for consumers after privatization.

Another argument against privatization is that it has negative effects on wealth distribution. The perception is that the rich profit far more than the poor, even when means aimed at evenly spreading the gains, like voucher schemes, are used. Birdsall and Nellis (2002) provide an overview of the available studies and conclude that most privatization programs seem to have worsened the distribution of income and assets, at least in the short run. In addition, they find that this is more evident in transition countries than in Latin America and less clear for utilities, where the poor have tended to benefit from greater access. McKenzie and Mookherjee (2002) look at the distributive impact of privatization in Latin America. They conclude that overall, the popular perception that privatization adds to inequality, cannot be confirmed.

While the welfare effects arguments of the opponents does not seem to have empirical founding the argument of worsening income distributions is a valid one, at least in the short run.

Next an overview of previous empirical research is given.

2.1.3 Empirical evidence

(16)

and Sachs, 2001). Boardman and Vining (1992) find that where competition is appropriate private ownership is preferable from an efficiency perspective. Furthermore Megginson et al. (1994) find significant increases in profitability, output per employee, capital spending and total employment for firms following privatization. Frydman, Gray, Hessel and Papaczynski (1997) compare the performance of state firms with privatized firms operating under reasonably similar conditions. However they supplement this comparison by examining the relative performance of the privatized and state firms in the period before the former were privatized. They too find that private ownership dramatically improves corporate performance.

Furthermore, Frydman et al. suggest that privatization is most important when the environment of firms is uncertain. Dewenter and Malatesta (2001) find for a sample spanning twenty years and several business cycles that government-owned firms are significantly less profitable than privately owned firms. They suggest that governments efficiently restructure before privatization. This however is already part of the process of privatization. Furthermore Zinnes et al. (2001) find that it takes more than change of name in order to increase performance as they find that so-called change-of-title does not have a significant impact on performance and therefore cannot conclude that change-of-title alone does lead to performance increases. Nellis (1999) provides us with a critical note on the results of others that privatization is always required and solely has positive outcomes, nevertheless he also concludes that privatization remains the generally preferred course of action, but stresses that a willing and capable government is essential. Good overviews of papers that confirm the improvement on firm performance after privatization are written by D’Souza and Megginson (1999), Megginson and Netter (2001), Kikeri and Nellis (2002) and Kikeri and Kolo (2005).

2.1.4 Macroeconomic evidence

(17)

Barnett (2000) also finds a positive relation between privatization and economic growth for a sample of 18 countries of which 12 are developing countries and 8 transitional. He finds that privatization of 1 percent of GDP is associated with an increase in the real growth rate of 0.5 percent in the same year and 0.4 percent in one year later. When the transitional countries are excluded from the sample these figures increase to 1.1 and 0.8 percent respectively. However, this evidence is not sufficient to establish causality.

The empirical analysis of Barnett (2000) is also used in Davis, Ossowski, Richardson and Barnett (2000). They suggest that privatization serves as a proxy for a range of structural measures, caused by a change of regime. Furthermore they argue that the positive effects of privatization are mainly caused by efficiency gains as they do not find a strong relationship between privatization and investment. With respect to unemployment rates they find decreasing numbers immediately as well as in the longer term. But they state it is difficult to isolate the effect of privatization here and the results are likely to be influenced by the effects of other policies. Therefore, it might be the case that the positive effects they found are caused by labor shedding as Kleinknecht (1998) finds that this might be advantageous in the short run. Labor shedding is a measure that is easier to implement than to innovate or invest and yields immediate results.

However, Kikeri and Nellis (2004) find that firms that privatized shed less than the economy as a whole. They suggest that these layoffs are a result of weak economic conditions rather than privatization. As a result they conclude that privatization does not increase unemployment and wage differentials.

(18)

deregulation and simplifying procedures for starting a business. McKenzie and Mookherjee (2002) emphasize that regulatory institutions need to be designed in order to ensure that: prices are kept low, firms operate under competitive pressure and are induced to innovate and keep costs low, and requirements are set for service expansion, quality, and access.

Secondly, there are the linkages to financial sector reforms. Dealing with issues of bankruptcy and protection of minority shareholders and tackling bad debt problems can significantly promote competition. Kikeri and Nellis (2004) find that the absence of these reforms lead to problematic privatizations (Czech Republic), while when these reforms are implemented, privatization programs run much more smoothly (for instance in Estonia and Hungary). Privatization of banks might be part of the solution. But good policy, monitoring and enforcement are essential elements here.

Regarding the development of regulatory frameworks, Kikeri and Nellis (2004) find that the key elements are the need for coherent policies, transparency and public disclosure, predictability in the rules of the game, a proper balance between autonomy and accountability and adequate institutional capacity. They argue that this is one of the most challenging aspects of privatization since changing ownership takes much less time than developing regulatory capacity and political interference occurs at many levels. In addition, McKenzie and Mookherjee (2002) state that the employment impact needs to be cushioned by funding severance packages, unemployment benefits, retraining, and job search assistance for the laid-off employees.

(19)

While at firm level the conclusion is straightforward that privatization is beneficial at a macro level the results are mixed. Two of the four4 researches mentioned above find positive effects of privatization, one finds negative results and one is unable to find a significant relation. Some important notes need to be made. First, the two papers that find positive effects make use of samples of developing countries and make use of lags5. In addition the samples they use are relatively small (35 and 20 countries respectively) compared with the sample of Cook and Uchida (2001) (63 countries). Considering Cook and Uchida (2001), who find a negative relation, do not make use of time lags and note that their results are sensitive combinations of control variables sample composition. Furthermore they argue that weaknesses in competition and the regulation of competition might explain their results.

Having provided this investigation of the relevant literature it is hypothesized that privatization increases efficiency at firm level. At country level the cumulative firm effects plus the positive short and long run fiscal effects and long run employment effects make that privatization enhances macroeconomic growth. This prediction can be summarized in the first hypothesis that remains unchanged:

Hypothesis 1. Privatization is positively related with economic growth.

Theoretically it is suggested that privatization is most important when the environment of firms is uncertain. This and the efficiency effects lead to the hypothesis that privatization is more beneficial for developing countries. Empirically, two papers making use of samples of developing countries find positive lagged effects of privatization on growth, while another one also making use of a sample of development countries that finds a negative relation does not make use of lags. The absence of time lags might explain this negative relation. Furthermore, a paper making use of OECD countries is unable to find a significant relation. Although these results come from different investigations and need to be compared carefully they do add to the theoretical expectation that privatization is more beneficial for developing countries. Consequently, my second hypothesis is as follows:

Hypothesis 2. Economic development weakens the relation between privatization and

economic growth.

4

The research of Davis et al. (2000) and Barnett (2000) are seen as one here as the paper of the latter provides the basis of the former.

5

(20)

Some additional literature needs to be evaluated to theoretical found the third hypothesis that considers the influences of the economic system of a country to the relation between privatization and growth.

2.2 Economic system

Historically dividing the world into different economic systems was based upon the differentiation between market economies and planned economies but due to the fall of the USSR the latter has, with a few exceptions, disappeared. In the past decade research has been focused on ways to differentiate between different forms of the former, i.e. market economies. One theoretical foundation for this research has been laid by Hall and Soskice (2001) with their varieties of capitalism. In this model economies can be compared by looking at the way in which firms cope with coordination problems they face in several spheres. Differences in way of resolving these problems are caused by several factors. First there is the role of institution and organizations and their capacities regarding exchange of information, monitoring and discipline in case of defection in relation to firms. Secondly the role of culture, informal rules and history cause firms within a economy to choose specific directions, that otherwise would not have been chosen. Furthermore, institutional infrastructure offer firms certain opportunities that lead to specific corporate strategies. Finally complementarities make that institutional practices are distributed in such a way that one increases return of another and therefore are expected not to be distributed randomly but clustered.

(21)

are all among the large countries from the OECD. Unfortunately this model cannot be extended to other economies in the world as the analysis of Hall and Soskice starts with the role of institutions thereby assuming that these institutions exist and are functioning at least in some way. However in some countries it might be the case that the mere problem is that such institutions do not exist or when existing are not functioning properly.

Nevertheless the work of Hall and Soskice provides a basic framework that can be used in this research. According to Thatcher (2004), at this moment several categorizations of capitalism exist, however most involve the same countries and use a similar classification.

Making use of the theory of Hall and Soskice it is hypothesized that SOEs of liberal market economies are more similar to private organizations than the SOEs from coordinated market economies. Therefore privatization of SOEs of liberal market economies is assumed to be less beneficial. This leads to the third and final hypothesis:

Hypothesis 3. The economic system of a country influences the relation between

privatization and growth, where in a more liberal economic system the effect is smaller.

To be able to investigate the effects of privatization on growth a model needs to be developed. It is important to consider the relevant literature as well as the empirical research on the most important growth explaining variables.

2.3 Theory on economic growth

In this research the effects of privatization in a country on economic growth are investigated. The theory on economic growth is very extensive. However a short overview is in place here since it helps to determine which variables are important in explaining economic growth. The model employed used in this research is based on a Barro growth model, first employed by Barro (1991)6. The Barro growth model is based on the Solow growth model, where output is a function of capital, labor and technology7. Mankiw, Romer and Weil (1992) find that saving population growth affect income as indicated by the Solow growth model and these two variables can explain more than half of the cross-country variation in income per capita. When the expand the Solow growth model by including both physical and human capital the explaining power of the model increases to eighty percent.

6

This model is later extended in Barro (1996 and 2000).

7

(22)

Barro (1991) finds evidence for the hypotheses of the Solow growth model and shows that growth, using the growth rate of per capita GDP as a proxy, is negatively related to initial level of per capita GDP. In addition, the growth rate of per capita GDP is positively related to the starting amount of human capital. He proxies human capital by school enrolment rates. Related, countries with high human capital have higher ratios of investment to GDP. The ratio of government consumption expenditure to GDP is found to be negatively related to growth. Furthermore, measures of political instability and price distortions are also found to be negatively related with growth. In this research per capita GDP growth is also used as dependent variable.

In addition to Barro (1991), there has been done extensive research in order to determine what variables are important in determining economic growth. Sala-i-Martin (1997) reports about 60 variables that turn out to be at least in some way related to growth. He points out that by simply “trying” a certain variable and putting it in a regression basically any variable could be related to economic growth. The key issue here is to point out the variables that are truly related to

explaining economic growth.

Levine and Revelt (1992) find that beside the investment share of GDP, the initial level of GDP, school enrolment rates and population growth rates no other variables are robustly correlated with growth. Sala-i-Martin (1997) uses an other measure than Levine and Revelt (1992) and runs four millions regressions to test which variables are strongly related to growth. Similar to Levine and Revelt (1992) he keeps the variables (i) initial GDP (proxied by log of GDP in 1960), (ii) human capital (proxied by primary school enrolment in 1960) and (iii) life expectancy in 1960 as fixed in his regressions as they already are empirically associated with growth. Doppelhofer, Miller and Sala-i-Martin (2000) use a Bayesian approach to determine which variables have strong explanatory power in cross-country growth regressions. This method relaxes some of the assumptions from Levine and Revelt (1992). Both Sala-i-Martin (1997) and Doppelhofer et al. (2000) find that regional factors like Latin American and Sub-Sahara African dummies and religious factors, e.g. fraction Muslim, fraction of Confucians and fraction of protestants are strongly related to growth and help explain cross-country variance.

(23)

3. Methodology

This study applies a cross section time-series, or panel data analysis to examine the relation between privatization and economic growth and, more specifically, the influences of economic development and system on this relation. Furthermore, this study draws on the results of prior studies that have investigated and identified various variables that explain economic growth. These studies, as well as this one, make use of cross-country data, which involves a number of conceptual and statistical issues. This research includes a all countries for which comparable data could be found. These are essentially very different from each other. Here it is argued, in accordance with Barro (2000), that it is not possible to perform an accurate empirical evaluation of growth implications from factors, such as in this case privatization, using the data of only a few countries.

One drawback here is, that, in principle, regression analysis requires observations that are drawn from a distinct population. Therefore including fundamentally different countries into a research may result in statistical bias, (Cook and Uchida, 2001). Furthermore, concerning this research specifically, is the missing data issue. Due to its nature privatization is a policy instrument that is not implemented on a yearly basis. This might be caused by economic circumstances or political conflicts. In that case the missing values are not random among the population and this may inflict bias on the distribution of the sample.

An important conceptual problem witch cross country analysis is, that the regression itself does not provide the complete answer to any causal relationships between policy variables and growth (Cook and Uchida, 2001). Contrarily, a time series analysis might not be able to find answers because the tests are too strong in finding systematic relationships. The effects on a single year might be too small to establish a relation. This might be, along with data insufficiencies, why prior studies made use of cross country regression analysis.

Yet the large data set used in this study provides, together with the potential explaining powers of this analysis, solid grounds to carry on and perform a growth regression analysis using panel data. The analysis can be conducted by applying linear ordinary least squares (OLS) regressions. The growth regression that is used as a basis in both regressions takes the form of:

(24)

where Y is the GDP per capita growth rate,

ε

the error term and xh a set of growth explaining

variables that are commonly used in the literature.

To investigate influences of economic development and system on the relation between privatization and growth a so-called interaction effect is included. To be able to investigate the influences of economic system the second regression the theory of Hall and Soskice (2001) is used.

3.1 Economic development

The first regression is used to test for the influences of economic development on the relation between privatization and growth. This can be investigated by making use of the so-called interaction effect. A basic interaction regression takes the form of:

Equation 2 Y =

β

_kx_k +

β

_lx_l +

β

_m(x_k×x_l)+

ε

where Y is the dependent variable, xk andxl are independent variables and

ε

the error term. The

effect of xkdepends on the value of xl and is grasped by the parameter

β

m. Here the assumption is

that the effect of privatization depends on the level of economic development, therefore xk is a

variable that represents privatization and xl a variable that represents economic development.

Equation 1 and 2 can be transformed into the following equation:

Equation 3 Y_ij =

β

_h_,_ijx_h_,_ij +

β

_k_,_ijx_k_,_ij +

β

_l_,_ijx_l_,_ij +

β

_m_,_ij(x_k_,_ij×x_l_,_ij)+

ε

where Y is the GDP per capita growth rate, xi are a number of growth explaining variables, xk a

variable that represents privatization and xl a variable representing economic growth. Term

ε

is

the error term.

A potential problem relating to this model is that economic growth is related to economic development. This would lead to correlation between one or more growth explaining variables and the variable representing economic development. More on this follows in the variable section 3.3 below.

3.2 Economic system

(25)

variable representing privatization in equation 3 is multiplied by several dummy variables each representing a different economic system. Such a regression takes the form of:

Equation 4

ε

β

+ + + + + + = _h_ij _h_ij _ij _k_ij _ij _k_ij _ij _k_ij _n_ij _k_ij _n ij x x D x D x D x D Y _, _, ₁_, _, * ₁ ₂_, _, * ₂ ₃_, _, * ₃ .... _, _, *

where Y again is GDP per capita growth. Xh are a number of growth explaining variables, xk the

variable representing privatization and D1,2,3,..,n are dummy variables for different economic

systems. Furthermore, term

ε

is the error term.

3.3 Variables and data

The following section explains variables that are included in the research and why certain proxies are chosen. As stated before the dependent variable in this research is GDP per capita growth as this variable is a widely accepted proxy for economic growth.

3.3.1 Growth explaining variables

Based on the literature the following growth explaining variables are included in the regressions, since they have found to be robustly and strongly related to growth. They are: (i) percentage investment share of GDP, (ii) percentage population growth rate, (iii) life expectancy as a proxy for human capital and (iv) log of per GDP capita in US dollars of the previous year as a proxy for initial wealth.

Contrary to prior studies, log of GDP per capita in 1960 is not used here, because the data here is a cross section time-series. Log is chosen because makes the data to be more evenly distributed, which leads to a more reliable estimation8.

Similar to Sala-i-Martin (1997) life expectancy is included as a proxy for human capital. The data for this variable is not available for every point in time. Linear interpolation is used to fill these missing data points. As the value of this variable does not alter much over time, this is considered to be a reliable method. Including school enrollment rates as a proxy for human capital is not considered an option here. The available data is small and the data available varies greatly over time. Filling the missing data points by either linear interpolation or out of sample prediction would result in an unreliable variable that lowers the explanatory power of the model.

8

(26)

Growth literature (Barro, 1992 and Barro and Sala-i-Martin, 1995, among others) indicates that investment share of GDP and life expectancy are expected to be positively related to growth, while population growth and initial income are expected to be negatively related to growth.

3.3.2 Privatization variable

Crucial to this research is the variable that represents privatization and the main question is how to measure privatization for a given country in a given year. A measure that has proven its functionality is cumulative proceeds from privatization as a share of GDP. Both Plane (1997) and Cook and Uchida (2001) make use of this measure. However Plane (1997) used the cumulative proceeds from privatization during the period 1988-1992 as a share of GDP in 1990 and turns this into a dummy. Cook and Uchida (2001) base their measure on the method of Plane (1997) but instead of GDP in 1990 they use average GDP during the period 1988-1992 as a weight. The measure used in this research is quite similar. The privatization variable used here is constructed by taking log values of cumulative proceeds in millions of US dollars from privatization per year as a share of GDP in millions of US dollars of the corresponding year. Log values are again chosen to improve the distribution of the data. Note here that, due the small privatization values, log values are negative. Nevertheless a positive coefficient in the regression indicates a positive relation and vice versa9.

An issue with using this measure could be that privatization variable reflects other policy and structural reforms, especially for developing countries. This problem is overcome by making use of panel data analysis. This makes it possible to isolate the effect of privatization proceeds on the economic growth by looking at a direct causal effect.

3.3.3 Economic development variable

The term economic development is difficult to define and can be interpreted in many ways. For instance, Ray (1998) defines economic development as “a concept embodying, not just income

and its growth, but also achievements on other fronts: reductions in infant mortality, higher life expectancy, advances in literacy rates, widespread access to medical and health services and so on” (p. 43). What is meant in this research with economic development is to provide a scale on which countries can be compared considering their state of development towards one another to

9

(27)

see whether this influences the relation between privatization and economic growth. This level of development is, among others, determined by factors such as efficiency, productivity, research and development, and levels of physical and human capital. From this line of reasoning it follows that the level of development determines the performance of a country. A measurement of performance is output and to scale this performance among a set of countries, output per capita is a useful measure even though it is an incomplete indicator for development in the broader sense. Therefore in this research the total value of all final goods and services produced within a country in a given year, divided by the average population of that country for the same year, or GDP per capita in US dollars is used as a proxy to measure and scale economic development.

3.3.4 Regional dummies

Additionally, to test for the first hypothesis the sample is divided into seven separate groups according to their more general characteristics. For each group a dummy variable is included in the regression. The countries and their group number can be found in appendix 1a.

Group 1: Europe

The countries in this group are considered to be among the most developed in this world. They share similar characteristics relating to free market systems and market efficiency. Furthermore governments are well developed and corruption is low. Some formerly transition countries are added to this group as these are considered to be at the end of the transition path, have joined or are in the process of joining the European Union and are further catching up to the richer nations.

Group 2: Transition countries

These economies of the former Soviet Union10 can be categorized into three groups. First there the Central and Eastern European economies (CEE)11, secondly the Baltics and third the economies in the Commonwealth of Independent States (CIS). Formerly these planned economies were characterized by a sellers’ market that ignored the demand side and lead to non-existence of pricing systems and competition. Furthermore firms were large, state-owned and vertically integrated and did not strive for profit maximization (IMF, 2000). In the early 1990s these countries entered the so-called transition process in order to bring them to the level of the

10

In this research the Asian transition countries are not considered here as they are allocated to other subgroups.

11

(28)

advanced industrialized economies and after a initial drop the output of the transition countries returned to the 1989 level at the end of the twentieth century.

Group 3: Newly Industrialized Countries (NICs)

Originally this group consisted of four Asian countries, namely Hong Kong, South Korea, Singapore and Taiwan but currently also countries like Mexico, the Philippines, Thailand, Malaysia, China, India and South Africa are said to belong to this group. These mostly Asian and formerly poor countries share some different characteristics from other economies in this continent. The main difference is that these countries have adopted outward-looking development strategies. This strategy involves government policies that keep markets open, maintain exchange rates keeping export prices at a competitive level whereby overall government interference is kept at a minimum (Husted and Melvin, 2001). They all have succeeded in developing particular sectors and industries that are competitive on world markets and thereby increased income levels significantly.(Goldstein, 2004).

Group 4: Latin America and the Caribbean

Even though this is a fairly large region to be quantified as one economic system there are several important features that unite the countries of Latin America and the Caribbean. The most important factor of these is history. Traditionally these countries have been united by language, religion, culture, bureaucratic outlook and they became independent in more or less the same period making them share the same experience of independence, which is based on the nineteenth-century liberalism of the European countries that they once belonged to. Furthermore they pursued similar economic policies (Cardoso and Helwege, 1999). The Latin American countries have for a long time been characterized by import substitution that led to high growth rates prior the 1980s, but then faced against them. The overvalued exchange rates due to protection caused a reduction in export growth and the import substitution made the industry grow at the expense of agriculture. In addition unemployment rose and large public deficits emerged, which led to high inflation and a debt crisis in the 1980s.

(29)

Group 5: Asia

This group exists of all Asian countries except the NICs and countries from the Middle East. Again colonial rule left his marks on countries in East and South Asia and the Pacific (Reynolds, 1986). The main difference with the NICs is that these countries did not adopt outward-looking development strategies but but kept to their import substitution policies. These mostly developing economies are characterized by slow economic growth, low efficiency and unemployment.

Group 6: Sub Sahara Africa

Like the Latin American and Caribbean group this group is also fairly large but with relatively equal characteristics. Another similarity with this group is that like Latin America and the Caribbean Africa has long been under colonial rule. However, in Africa new territorial boundaries, political relationships, social strata and economic structures were created by the European colonizers (Lewis 1998). Especially the first creation led to an important extra development challenge after the decolonization wave in 1960, i.e. the challenge of nation building. Furthermore colonial rule left weak economic and political foundations to most countries (Lewis 1998). The following decades were ones of conflict and crisis. At the end of the 1980s decade a wave of democratization followed. Under Conditionality economic reform programs were pushed through in trade for financial aid, where forced privatization became a popular in the 1990s. Today this continent houses most of the least developed countries in the world and most African countries remain highly dependent upon a narrow range of primary commodities for export earnings.

Group 7: North Africa and Middle East

(30)

3.3.5 Economic system

To test for the third hypothesis a regression analysis is done based on the theory of Hall and Soskice (2001). As stated in section two they differentiate three varieties of capitalism. The first are the LMEs where firms coordinate their activities primarily via hierarchies and competitive market arrangements and supply and demand in competitive markets are very important issues. Next are the CMEs. Here firms depend more heavily on non-market relationships to coordinate their activities with other actors and to construct their core competencies. Strategic interaction and networking are important issues with CMEs. Third are the Mediterranean economies, which is more a mixed group with capacities for non-market coordination in the sphere of corporate finance but more liberal arrangements in the sphere of labor relations caused by a large agrarian sector and a history of extensive state intervention. A dummy variable is created for each group.

3.3.6 Data

Data for the independent variable GDP growth per capita as well as the independent variables population growth rate and GDP per capita is taken from the World Bank GenderStats database. The data for the independent variable investment share of GDP is obtained from the Center for International Comparisons at the University of Pennsylvania (CICUP). The privatization proceeds data are obtained from the Privatization Database of the World Bank and the Privatization Barometer of the Fondazione IRI and the FEEM. The privatization proceeds in these databases are defined to include all monetary receipts to the government resulting from transactions involving partial and full divestitures, concessions, management contracts, and leases. The data for a number of European developing countries of the Privatization Database come from the Privatization Barometer. Therefore the data of the other European countries from the Privatization Barometer is used to form a larger database, including more countries.

(31)

present and for the Financial Market Trends issues from the OECD for the period 1990-2001. Data on investment share of GDP is available until 2000. Data on all other variables is available beyond the periods named above. Therefore, the time span covered in this research is 1990 to 2000 simply due to data restrictions.

3.3.7 Sample

(32)

4. Empirical analysis

In this section the hypotheses are empirically analyzed. The set up is as follows. First the data is analyzed by an inspection of the descriptive statistics. Next a visual inspection of the data is performed to check for any initial relations. Since this research handles macroeconomic time series data, the issues of multicollinearity and stationarity are important and are therefore elaborated further after the visual inspection. I then proceed with an analysis of which regression model should be used, after which the most appropriate is used to perform a regression analysis to test the hypotheses. Since two samples are used this process is carried out twice. Afterwards a summary of the results is given.

4.1 Cross country time series analysis.

The relation between privatization and economic growth and the effects of economic development are investigated making use of a panel data analysis.

The countries included in the sample that is used for this analysis are stated in Appendix 1. This table states the countries alphabetically by number (1 to 112). In addition behind the country name the group number is stated in which the country is allocated. In Table 4.1 the shares of the different groups in the sample can be found.

Table 4.1. group shares

Group number and name Nr. Countries Percentage

1. Europe 21 19

2. Transition Countries 17 15

3. NICs 7 6

4. Latin America & Caribbean 21 19

5. Asia 8 7

6. Sub Saharan Africa 32 29 7. North Africa & Middle East 6 5

(33)

Table 4.2. Descriptive statistics Variable Per capita GDP growth Investment share of GDP Average population growth Log of per capita GDP Life expectancy Yearly proceeds from privatization Interaction N 1753 1346 1790 1768 1792 858 858 Minimum -0.44 0.01 0.06 4.31 36.03 -14.12 -122.68 Maximum 0.35 0.47 0.11 10.31 80.11 -1.20 -7.69 Mean 0.01 0.14 0.01 7.29 64.84 -5.72 -43.99 SD 0.06 0.07 0.01 1.42 10.72 1.88 16.77 Skewness -1.97 0.72 -0.19 0.28 -0.86 -0.63 -0.93 Kurtosis 14.68 3.78 4.95 2.27 2.61 3.68 4.24 Jarque-Bera 11108.26 149.20 296.32 61.88 233.49 73.38 179.63 Probability 0.000 0.000 0.000 0.000 0.000 0.000 0.000

(34)

The next step is a visual inspection. This can be done by plotting each variable separately on the dependent variable per capita GDP growth. The results are stated in appendix 3. As can be seen no clear cut relations can be found. This also might be an indication that fixed country effects are present.

4.1.1 Multicollinearity

To detect for multicollinearity a correlation matrix use which can be found in table 4.3.

Table 4.3 Correlation Matrix.

Per capita GDP growth Invest- ment share of GDP Popu-lation growth Life expec tancy Log of per capita GDP Log of yearly proc. of priv. Inter- action Per capita GDP growth 1.00 0.14 -0.05 0.11 0.04 0.10 0.07 Investment share of GDP 1.00 -0.35 0.62 0.66 -0.03 -0.37 Population growth 1.00 -0.58 -0.55 -0.18 0.13 Life expectancy 1.00 0.83 0.04 -0.39 Log of per capita GDP 1.00 0.06 -0.48 Log of yearly proceeds of

privatization 1.00 0.83

Interaction 1.00

A common rule of thumb is that when the correlation coefficients between two explanatory variables have higher values than 0.8 a strong linear association is to be expected. (Carter Hill, Griffiths, Judge, 2001) As can be seen in table 4.3 a correlation between life expectancy and log of per capita GDP of 0.83 and a correlation between the log of yearly proceeds from privatization and the interaction variable of 0.83 could signify collinearity.

However a correlation matrix only examines pairwise correlation and therefore a more effective test is to estimate so-called “auxiliary regressions”. In an auxiliary regression the left-hand-side variable is the suspected explanatory variable and the right hand side are all the remaining explanatory variables. If the R squared of this model is high (a common rule of thumb is above 0.80) most of the variation in the suspected variable can be explained by variation in the other explanatory variables and multicollinearity is present.

(35)

the left-hand-side variable and has a R squared of 0.97. The conclusion here is straightforward and log of yearly proceeds from privatization and the interaction variable cannot be present in the same regression. This means that the relation between privatization on economic growth and the effect of economic development on this relation need to be tested with different regressions.

4.1.2 Stationarity and unit roots

A special issue with time-series data is stationarity. When two nonstationary variables are regressed on each other spurious results are obtained, i.e. misleading significant regression results. With time-series data stationarity can be checked for by testing for unit roots making use of the so-called Augmented Dickey-Fuller test. For panel data the Levin, Lin and Chu (LLC), and the Im, Pesaran and Shin (IPS) test can be used to test for stationarity by checking for unit roots. Both test for the null hypothesis that each cross section in the panel has non-stationary time series, but the alternative hypotheses used are different. LLC test the null against the alternative hypothesis that all cross sections’ time series are stationary, while with the IPS test the first order auto regressive coefficients are allowed to differ across cross sections in the alternative hypothesis. The different variables used in the analysis are all tested for unit roots. The results can be found in table 4.4.

Table 4.4. Results of unit root tests

Levin-Lin-Chu test Im-Pesaran-Shin test Variable Test statistic Prob. Test statistic Prob. Per capita GDP growth -18.8050 0.0000 -16.4422 0.0000 Investment share of GDP -15.5746 0.0000 -7.7016 0.0000 Population growth -9.6212 0.0000 -10.9317 0.0000 Log of per capita GDP -3.7789 0.0001 2.5300 0.9943 Life expectancy -14.4493 0.0000 -0.9184 0.1792 Log of yearly proceeds of privatization 14.2629 1.0000 -3.1694 0.0008 Interaction 18.9943 1.0000 -2.7523 0.0030

As can be seen in table 4.4 the LLC test indicates that both the privatization variable and the interaction variable have unit roots and the IPS test indicates unit roots for log of per capita GDP and life expectancy. There are however some issues concerning these tests and stationarity and panel data in general.

(36)

lead to the rejection of a series in a regression due to failure to reject the null hypothesis of the LLC test, while this series might be of useful explanatory power in the regression. Second, according to Maddala and Wu (1999) an important issue with the IPS test is that this test assumes that the number of periods (T) is the same for all cross sections (N), and thus only balanced panel data is considered. Using unbalanced data brings difficulties. An important issue involved with both tests is that they assume that N and T go to infinity, which makes these tests more suitable for macro panels where both the number of units and time series are large (Hadri and Larsson, 2004).

Furthermore, some general issues with stationarity and panel data exist. As indicated by Libanio (2005) most macroeconomic time-series are expected to have unit roots and are therefore not mean reverting. Furthermore Smith (2001) concludes that in case of, for instance, forecasting or estimating long-run parameters making use of standard pooled estimators like fixed effects may perform well. Maddala and Wu (1999) conclude that many tests for stationarity are not likely to be meaningful in practice as it drifts the analysis away from the questions that are being answered. In addition, Libiano (2005) states that many scientists argue that the implications of the presence of unit roots in this models are minimal and concludes that ‘There seems to be no consensus

about the most appropriate methodologies to perform unit root tests; no consensus about the theoretical importance of the concept of unit roots and its implications for macroeconomic analysis.’ (p. 173).

(37)

4.2 Privatization and growth

Next a regression analysis is conducted to estimate the coefficients in the different models. Due to the likeliness of country specific effects a formal test is used to determine whether to use a fixed effects model or a common effects model. The fixed effects model allows the intercept parameter to vary across countries. The common effects model does not allow for country specific effects. The details on this test can be found in Appendix 5. As can be seen the test yields a test statistic of 2.397. This exceeds the critical value of 1.260. Therefore a fixed effects model is more appropriate.

4.2.1 Regression analysis

First, the relation between privatization and growth is investigated. After an investigation of the theory and empirical work I hypothesize this relation to be positive. Therefore the null hypothesis here is:

H0: Privatization has no or a negative effect on growth.

This is tested against the alternative hypothesis Ha that privatization has a positive effect on

growth.

To test this hypothesis the following regression model is constructed:

ε

β

+ + + + + + + + +

= D D D INV POP LE LCAP− LPRO

Y 1,1 1 1,2 2 ... 1,112 112 2 3 4 5 t 1 6

where subscripts 1 to 112 associated with β1 and D indicate the countries in the sample listed

alphabetically and D is the fixed country effect parameter, INV is investment share of GDP, POP population growth, LE life expectancy, LCAPt-1 log per capita GDP in period t -1, LPRO log of

yearly proceeds from privatization and

ε

is the error term. Note that due to the problem of multicollinearity the interaction variable is excluded from the regression. A separate regression will be run later on.

(38)

To test for autocorrelation (whether the error terms are uncorrelated), the Durbin-Watson (DW) statistic can be used, which has a value of 1.5012. The statistical package Eviews does not provide a critical value. Therefore lower and upper bounds are used here to determent the presence of autocorrelation. The lower and upper bound, for a regression with T=10 and a number terms on the right hand side including the error term of 6, are 0.243 and 2.822 respectively at the 5 percent significance level. The test for autocorrelation is inconclusive according to the bounds test as the DW statistic fall between the lower and upper bound. However, considering a DW statistic of (close to) 2 is desirable, these are rather extreme bounds and, here, the bounds test is inappropriate. Actually, with a DW statistic of 1.5012 autocorrelation is considered to be a serious problem. This is quite plausible since shocks in economic variables are not expected to work themselves out in one period. This has several consequences for the least squares estimator. First of all, the least squares estimator, even though unbiased, is no longer best. Second, hypothesis tests may be misleading because the formulas to calculate the standard errors are no longer correct.

The most common (and simple) solution is to use a first-order autoregressive model to estimate the relation. In this model the error depends on its own lagged value, plus another random component that is unrelated over time and has zero mean and constant variance.

Since such a model compares years, this has a significant decreasing effect on the sample size, which drops from 112 to 88 countries due to data limitations. A list with the countries that are dropped is given in appendix 6b. Here can be seen that most countries that are dropped are Sub Saharan countries. Appendix 6c states the complete new sample and the new group sizes. This sample will be used in all tests unless stated otherwise.

The results of the autoregressive model can be found in Appendix 7. A summary is given in table 4.4.

Table 4.4. Regression results

Independent Variable Coefficient Probability

ε _4.3575 _0.0000

Investment share of GDP 0.6713 0.0000 Population growth -0.9834 0.0030 Life expectancy 0.0166 0.0014 Log of per capita GDP (t-1) -0.7125 0.0000 Log of yearly proceeds from priv. -0.0008 0.2883 Auto regressive first order 0.7843 0.0000 Number of observations 511

(39)

As can be seen in table 4.4 the model as a whole is significant with an F value of 5.4403 and a significance of 0.0000. The signs of the growth explaining variables correspond to those predicted according to the literature. Initial per capita GDP has a negative and statistically significant coefficient in accordance with Mankiw, Romer and Weil (1992) and Barro and Sala-i-Martin (1995), population growth also has a negative coefficient like Barro and Sala-i-Sala-i-Martin (1995) find, although their coefficient is insignificant while here the coefficient is statistically significant. Life expectancy has a positive significant coefficient in accordance with Barro and Sala-i-Martin (1995) and the estimated coefficient of investment share of GDP is positive and statistically significant like was found in Levine and Renelt (1992) and Mankiw, Romer and Weil (1992).

The coefficient of the privatization variable has a negative sign. However, this is not significant. The significant parameters explain about 55 % of the variance in the residuals of the model, this is expected to be caused for a great part by the country specific intercepts because, as can be seen in appendix 7, the country specific parameters vary greatly among each other. Nevertheless the model as such does not find any relation between privatization and economic growth and the null hypothesis cannot be rejected.

4.2.2 Regional effects

The insignificant relation found above might be caused by regional differences. When these differences are large enough no linear relation can be found when the countries of al these regions are pooled together in one sample. To test for this seven dummies are created, each representing a region from table 1. These dummies are the multiplied by the privatization variable, to allow the effect of privatization on growth to differ between regions. The results can be found in appendix 7b and a summary is given here in table 4.5.

Table 4.5. Results regional effects

Variable Coefficient Probability

ε _4.2760 _0.0000

(40)

Privatization. Asia 0.0012 0.3321 Privatization. Sub Saharan Africa -0.0037 0.0146 Privatization. North Africa & Middle East -0.0049 0.0419 Auto regressive first order 0.7799 0.0000 Number of observations 511

F-statistic 5.1519

Probability F statistic 0.0000

R-squared 0.55

As can be seen here, the overall results improve only marginally. For most regions there is no significant relation. However some rather unsuspected findings can be found for the regions Sub Saharan Africa and North Africa and the Middle East with coefficients of -0.0037 and -0.0049 respectively. A significant negative association between privatization and growth is found here, which confirms the results of Cook and Uchida (2001).

An explanation for this negative relation might be that in trade for financial aid economic reform programs were pushed through, including forced privatization programs. Reaping the gains from privatization requires complementary institutional reforms to become fruitful as stated in section two. Developing countries often do not have the capabilities to successfully implement these reforms. It also might be the case that endogeneity plays an important part here. The initial and regional model only measure immediate effects. When a privatization program is implemented in countries during economic bad times an immediate positive effect may be difficult to find. In addition the short run negative employment effects might be greater than the positive fiscal effects causing a negative relation between privatization and growth.

4.2.3 Lagged effects

As stated in the literature section the only two immediate positive macroeconomic effects are the positive fiscal effects and negative unemployment effects. Where the positive fiscal effects are often small. It might take several years before all measures and people are in place and fully functioning and therefore the effects are visible. It is therefore far more likely that the effects of privatization on growth are lagged.

Privatization and economic growth in the 1990s A panel data approach

Privatization and economic growth in the

1990s

A panel data approach

J.J. Rodenburg

Privatization and economic growth in the

1990s

A panel data approach

J.J. Rodenburg

Abstract

Preface

Contents

1. Introduction

2. Theoretical background

3. Methodology

ε

β

β

β

ε

ε

β

β

β

β

β

ε

ε

ε

β

β

β

β

β

ε

4. Empirical analysis

ε

β

β

β

β

β

β

β

β

ε