THE IMPACT OF REGULATORY GOVERNANCE AND PRIVATISATION ON ELECTRICITY INDUSTRY GENERATION CAPACITY IN DEVELOPING COUNTRIES

Transcription

1 THE IMPACT OF REGULATORY GOVERNANCE AND PRIVATISATION ON ELECTRICITY INDUSTRY GENERATION CAPACITY IN DEVELOPING COUNTRIES By John Cubbin, City University and Jon Stern, London Business School ABSTRACT This paper assesses for 28 developing countries over the period whether the existence of a regulatory law and higher quality regulatory governance are significantly associated with superior electricity industry outcomes. The analysis draws on theoretical and empirical work on the impact of developing country telecommunications regulators. The empirical analysis concludes that, controlling for other relevant variables and allowing for country specific fixed effects, both a regulatory law and higher quality regulatory governance are positively and significantly associated with higher per capita generation capacity levels, controlling for privatisation and competition. In addition, this positive impact continues to increase for over 10 years as experience develops and regulatory reputation grows. The results are robust to alternative dynamic specifications, including estimates from error correction models, to the inclusion of country governance political risk indicators and to controlling for potential endogeneity biases. The paper concludes with a short discussion on causality in panel data modelling of governance models and the policy implications for regulatory reform. 1

2 1. Introduction Over the last years, a very large amount of attention has been given to the role of institutions in economic growth. This has, in large part, been driven by economic policy priorities such as how to develop effectively functioning market economies in Central and Eastern Europe and the former Soviet Union post-1989; and how to foster economic growth in lagging world regions such as Sub-Saharan Africa. In parallel, and partly in response, there have been major explorations of the role of institutions in the functioning of market economies. In addition, in recent years, there has also been a substantial empirical literature on the relative roles of institutions, policy, geography and trade openness on growth performance across countries. This literature places considerable weight on institutional quality as a major determinant of variations in long-term growth performance 1. In particular, Rodrik (2003) argues that there is a requirement for a cumulative process of institution building to ensure that growth does not run out of steam and that the economy remains resilient to shocks 2. The arguments above on aggregate growth apply with extra force to utility service industries. This is because not just are they highly capital intensive, but, in addition, because most of their assets are very long-lived and (in economic terms) sunk assets. Hence, an effective institutional framework is essential to sustain growth in output, efficiency and capacity for commercialised utility service industries such as electricity, telecommunications, water and similar - particularly if these industries have significant amounts of private investment (physical and/or financial). The standard institutional solution to handle these infrastructure industry issues is to introduce an independent regulatory agency, operating within a clearly defined legal framework 3. The regulatory agency is intended to provide the high quality institution which permits and fosters sustained growth in capacity and efficiency in the utility service industries particularly the network elements. Hence, whether or Work on this paper was supported by the research program on Industrial Organization Policy for Development at the Development Research Group of the World Bank, under the direction of Ioannis Kessides. We are grateful for helpful comments from seminar participants at the University of Cambridge and City University as well as from the Editor and three anonymous referees. The authors alone are, however, solely responsible for the analysis and the views expressed See Rodrik, Subramanian and Trebbi (2002) for a recent survey of the literature on studies of cross-country growth performance. Rodrik (2003), p.25 An independent regulatory agency is not the only way of providing the necessary institutional support either in theory or in practice See Domah, Pollitt and Stern (2002). In addition, an independent regulator may be combined with a high or a low degree of reliance on contracts and courts. There is a major issue of whether or not low income developing countries have the human and other resources to sustain independent regulatory agencies, particularly regulatory agencies with a significant degree of discretion. Nevertheless, an independent regulatory agency has become the standard recommended solution to the private investment problem for utilities in the same way as an independent central bank has become the standard recommended solution to handle commitment and time inconsistency problems in monetary policy. See Stern and Cubbin (2003). 2

3 not country X has a high or a low quality institution is determined primarily by the quality of governance of the regulatory agency (conditional on the governance quality for the country as a whole). As with the aggregate economy, developing countries with high quality regulatory agencies (as measured by their regulatory governance) should attract more investment on a sustained basis into their utility service industries and at a lower cost of capital, as well as having higher efficiency levels and growth rates in the regulated utilities 4. The perspective outlined above is at the heart of the recent literature on regulatory governance for utility service industries, particularly the literature that focuses on developing and transition economies. This perspective is set out in Levy and Spiller (1994) which draws explicitly on North (1990) as well as in a number of subsequent papers 5. However, there have also been cases where the apparent outcome regulatory (and electricity) reforms has disappointed. There have been many case studies and these can be very illuminating but do not allow reliable generalisations but, until the last 2-3 years, little formal econometric or other statistical testing. Until recently, however, there has been very little systematic empirical testing of the view and policy that improved regulatory governance increases investment or efficiency in the electricity industry. There have been a number of recent systematic empirical studies for telecommunications. In this paper, we carry out a similar exercise for electricity supply industries in developing countries. Specifically, we provide an econometric analysis of the relationship between the quality of regulatory governance and the level of generation capacity per capita for a sample of 28 Latin American, Caribbean, Asian and African countries over the period , taking account of privatisation and competition. The plan of the paper is as follows. In Section 2, we outline the underlying economic issues and the main institutional design considerations as well as related recent research. In Section 3, we outline our modelling approach. Section 4 discusses the main econometric results from static and dynamic models and Section 5 discusses issues of endogeneity and causality. Section 6 presents a discussion and concluding comments The problems for developing country governments over making credible commitments to support new investment in the presence of major fixed costs arises in other contexts besides utility regulation. A good example is export taxes for exportable cash crops. See McMillan (2001). We are grateful to the editor for this observation. See, inter alia, Smith (1997), Stern and Holder (1999), Noll (2001). See Cubbin & Stern (2004) for a much fuller version of the,paper, particularly of the data and tables of results. 3

4 2. Underlying Economic Issues, Institutional Design and Implications for Empirical Analysis The main issue on which we focus is the inability of governments to make credible and binding commitments about utility pricing to sustain private investment while retaining decision-making powers over these issues. The discussion of utility service regulation concentrates on commercialised utilities facing genuine budget constraints, particularly where private investment and/or private finance is important. The focus of the discussion (and of our empirical work) is on regulatory governance (e.g. autonomy, accountability, etc) rather than on regulatory content (e.g. methods of price, investment and related aspects of regulation) 7. The underlying economic issue for utility regulation as for monetary policy is that governments, particularly at certain times, have a strong incentive to behave in a short-sighted and populist manner that reduces welfare summed over a medium to long-term period. 2.1 Output Measures for Utility Regulatory Agencies For utility service industries, there are two main output measures for measures for utility regulation. These are: (i) (ii) the level and rate of growth of technical efficiency and productivity (and of quality of service); and the level of capacity. In this paper, we focus on developing countries and, in particular on capacity levels. Hence, our estimates provide a test of the key policy objective of the World Bank and many of the countries in the sample. They have consistently cited significantly higher investment (and private investment) as the single most important reason for the promotion of independent regulatory agencies in electricity and similar utility service industries 8. In consequence, on this hypothesis, we would predict: (a) (b) sizeable increases in investment flows (domestic and foreign) developing country electricity industries following the establishment of an regulatory agency; larger increases with higher quality regulatory governance; and 7 8 We looked, in passing, at methods of price/profits regulation in our empirical work but this issue was a subsidiary concern for this paper. See Section 4 for the results. The World Bank s 1994 World Development Report Infrastructure for Development is a good example. See Chapter 3. 4

5 (c) larger impacts as the regulatory agency gains experience and reputation Previous Literature Our empirical work adopts and extends the fixed effects panel data modelling that has been used in the literature on the impact of regulation on telecom outcomes. See, for instance, Fink, Mattoo and Rathindran (2003), Wallsten (2002) and Gutierrez (2003). The approach of Gutierrez (2003) is particularly relevant to this paper. He constructs a regulatory governance index for his sample of 22 Latin American and Caribbean countries. This 7-element index (derived from the Stern-Holder typology) is calculated from examination of each country s telecom laws and changes in the laws. In our model for electricity outcomes, we adopt a similar approach and make use of a snapshot 4-element index as one of our regulatory variables. Gutierrez (2003) finds statistically and positive direct effects of his regulatory index both on tele-density and on efficiency. This result occurs both in static and dynamic models and after testing for the endogeneity of regulation 9. The estimated effect of a 1-point increase in the index on mainlines per 100 inhabitants varies somewhat depending on the precise model specification but is, in general, of the order of 20%. For electricity, there are so far only a very few and very preliminary empirical studies of the impact of regulation e.g. Zhang, Kirkpatrick and Parker (2002) and a part of Pargal (2003). Like this paper, they also concentrate on generation capacity but find only weak effects (if any) of regulation. Their studies also have major problems in disentangling the effects of regulation per se from those of privatisation and liberalisation. However, the studies are much more preliminary than those for telecoms, particularly in regulatory and other data terms. In this paper, we have access to better data on regulatory governance and its variation across countries. However, again, data constraints confine us to estimating capacity models for generation rather than transmission, distribution, sales or commercial losses 10. Regulatory issues are, of course, only one aspect of electricity industry reform. For a comprehensive discussion of electricity reform in developing countries, see Jamasb et al (2005). 10 Cubbin and Stern (2004) report estimates of the impact of regulation on technical losses. 5

6 3. Economic Rationale, Model Specification and Modelling Issues The modelling work reported in this paper is concerned with whether better regulatory governance in developing countries increases rated generation capacity per capita. 3.1 Economic Rationale In developing countries, the introduction of explicit regulation is to focus the policy of the electricity industry on providing sufficient supplies and that, typically, means increasing investment and capacity. In some cases, this has been done by harnessing the forces of private ownership and domestic or foreign private investment. In others, it has to provide a workable financial framework within which the electricity industry could develop by loosening the ties with government. This can take place, for example, by a country enacting an electricity law giving various powers and duties to a Ministry or independent regulator. Such changes can also increase public investment in infrastructure industries e.g. by requiring state owned electricity companies to operate in a more commercial way and hence allowing access to these companies to private debt finance for investment at a reasonable cost of capital. Investment is encouraged once effective regulation is available to support a workable financial framework. If the electricity industry is in private ownership the owners have the prospect of earning a reasonable return on their investment; if publicly owned, the industry can become independent of tax revenue or continually increasing loans. In addition, the existence of an effective regulatory framework can also encourage the growth of private investment and/or private finance within state systems, as has been happening in recent years in India and China. In an unconstrained market economy, per capita generation capacity will adjust to the level of demand, which will depend upon the level of per capita income, the price of electricity, and environmental factors such as climate. The price of electricity will be determined in part by the efficiency of the sector. The latter may depend upon regulatory factors, but also availability of energy sources such as hydro, gas, oil, and coal. However, many developing countries with a traditional, vertically integrated and state-owned electricity sector will be constrained not so much by market demand but by the availability of continuing subsidy. In constrained developing country electricity markets with implicit or explicit subsidies, capacity constraints arise because of either (a) inadequate government revenues for electricity investment or subsidy payments; and/or (b) insufficient revenue flows to support viable private investment or commercial debt obligations. Electricity generation models for unconstrained markets typically find per capita GDP to be the major determinant of electricity demand (and hence of generation capacity). We therefore include per capita GDP in our model as well as other control variables that have been found to be statistically significant in previous studies of developing country infrastructure industry infrastructure industry capital requirements eg the share of industry in value added, country debt levels and country economy-wide governance indicators. 6

7 We include in the model variables for electricity privatisation and competition and also variables on country governance quality. An effective regulatory framework can be expected to reduce the constraint on the operation of the market, increasing supply and moving the outcome closer to the market equilibrium. The better the governance of the regulator, the greater the expected increase in capacity and increase in electricity supply. 3.2 Model Specification The considerations above suggest a model where there is a long-run equilibrium capacity-output relationship for generation capacity for each country, varying by country. For developing counties, with supply constrained electricity, improved regulatory governance is expected to raise equilibrium generation capacity levels. The adjustment to the new equilibrium is very likely, however, take some time to achieve. This suggests a long-run static model of the following form, which is specified below in panel data format: Log(ELCAPPC) it = (a 0 + a i ) + a 1 log(gdppc) it + a 2 RegVar it + a 3 X it + v it (1) Where Log ELCAPPC is the log of per capita electricity generation capacity in Gigawatts; a 0 is a constant term; a i is a time-invariant country specific fixed effect GDPPC is real per capita national income in $US ; RegVar comprises one or more of the regulatory governance variables X is a vector of other potentially relevant sectoral and country level control variables and v it is an error term In all cases, the variables are defined for i = 1,, I countries over t = 1,., T time periods. Note: 1) The X vector of control variables for this equation might well include domestic fuel/hydro source availability and a variety of other country specific 11 Hence, GDP is on an exchange rate rather than a PPP basis. 7

8 economic and/or institutional variables. However, we expect that both of these will largely be captured by the country-specific fixed effects. Similar arguments apply to institutional/country governance effects since country rankings on these indicators tend to be relatively stable over year periods. 2) We also explore whether either (a) privatisation and/or (b) competition affect generation capacity growth. We investigate both direct and indirect effects (e.g. interactions between these variables and the regulatory variables). 3) On the basis of previous studies of electricity demand, we would expect a 1 to be close to but probably less than The equation above is a static representation of the model, which provides evidence on long-run equilibrium effects. We also consider some dynamic error correction models, which provide evidence on the adjustment time-path and separates short-run adjustment effects from long-run equilibrium effects. To ensure that our modelling yields estimates of supply responses, we confine the sample of countries to countries with unsatisfied demand for electricity throughout the period ie developing countries in Africa, Asia and Latin America only. We exclude both developed countries, and European Transition economies as both have significant planning margins and/or unutilised capacity for some if not all years of the period. 3.2 Data Our sample is of 28 developing countries. We have complete (or near-complete) generation capacity data on these countries for a 21 year period ( ). This gives us a longer panel than is usually available for such studies; this greatly reduces the econometric problems associated with short panels. There are, however, some missing observations so that it is an unbalanced panel. Of the 28 countries in the sample, 15 were in Latin America, 6 in the Caribbean, 4 were in Asia and 5 were in Africa. The list of countries includes large countries (e.g. Brazil and India), small countries (e.g. Jamaica); middle income countries (e.g. Chile and Mexico) and poor countries (e.g. Ethiopia and Sudan). The full list of countries and summary regulatory characteristics is listed in Appendix 1. It is worth noting that out of the 26 regulatory reforms listed, only 6 are pre The dependent variable in our regressions is per capita generation capacity by country and year. This is derived from the US Energy Information Agency data on generation capacity by country (GW) See Dahl and Roman (2004) Table 5 for a recent survey of electricity demand elasticities. 8

9 Graphs of the data are reproduced at Appendix 2. Interestingly, they are very different by country and there are some significant decreases (Nigeria and Nicaragua ) as well as large jump increases (Paraguay ). Generation capacity changes tend to be lumpy so that our dependent variable, does not obviously exhibit common or stable trends. (Note that the EIA series does not distinguish between publicly and privately owned generation capacity.) The key independent variables for this study are the regulatory variables available to us. There are data for each country on the existence (or absence) of: (i) (ii) (iii) (iv) an electricity or (energy) regulatory law; an autonomous or a Ministry regulator; licence fee or government budget regulatory funding; and free or mandatory civil service pay scales for regulatory staff. Each of these is measured by a 0/1 dummy. The dating of the switch from 0 to 1 on the appropriate variables (subsequently maintained at a constant level) is derived from the date of enactment of a primary electricity reform or regulatory law (except for cases where other information was available to provide a known, superior alternative). Hence, we can investigate the effect of age of the regulatory agency as well as its existence so that we can estimate alternative measures of the impact of regulation based on the age of the regulator 13. Given the time needed to establish a functioning regulatory entity, the start date for regulation is taken as the year following the enactment of the law. The regulatory variables in our index are all measures of formal attributes of regulation. Unfortunately, no comparable data is currently available on the informal, practical qualities of electricity regulation (eg transparency and quality of regulatory processes). The necessary omission of data on these characteristics may lead to potentially biased estimates and standard errors 14. In addition, unlike Gutierrez (2003), we have no time dimension on changes in formal governance attributes subsequent to the enactment of the primary electricity/energy regulatory law. The Domah data set is very suitable for a preliminary investigation of the impact of regulation but is not ideal. In particular, it suffers from an absence of data on the informal, practical aspects of regulation (e.g. length of tenure of regulatory agency heads or commissioners, etc. Although much of the regulatory activity took place in the last half of the data set, the earlier period is important in effectively establishing benchmark pre-reform levels of These regulatory data were collected in a 2001 study by Preetum Domah. See Domah, Pollitt, and Stern (2002) for full details. We are very grateful to Preetum Domah for permission to use the information from his survey in this paper. See Stern & Cubbin (2003) pp where preliminary simulation results based on the Stern-Holder data set suggest that omitting data on the informal, practical aspects of regulation can lead to coefficients being under-estimated by around 5-10% and a similar under-estimate of t-values. 9

10 generation capacity, and also in reducing some of the biases that can potentially arise in the use of short panels. However, 20.7% of the total number of country-sample years were years with an autonomous regulator and 31% with an electricity or energy regulatory law. By the end of the period, only 2 of the countries had not enacted an electricity law but there were 9 countries with a Ministry regulator operating under a law. 15. A key feature of the regulatory data is that the correlation between the four regulatory variables is, not surprisingly, very high. In addition, all of the countries with an autonomous regulator had an electricity law as did all the countries with licence fee funding 16. The matrix of correlation coefficients between the regulatory variables is as follows: Correlation Matrix of Regulatory Variables Licence Fee Funding Non Civil Service Pay Scales Electricity Law Autonomous Regulator Electricity Law 1 Licence Fee Funding Autonomous Regulator Non Civil Service Pay Scales This high level of collinearity between the regulatory variables presents estimation problems which we discuss in the next section. For privatisation and competition, we use the Henisz-Zellner-Guilen (HZG) (2004) electricity data 17. On privatisation, this data set provides information on the year in which all countries introduced: (a) minority privatisation of their electricity industries; (b) majority privatisation of their electricity industries and; (c) total privatisation of their electricity industries. The HZG data on competition include a variable for the year in which private firms were legally allowed to generate electricity for resale. But, this does not necessarily mean that such electricity sales were important or even present and one-half of the countries in our sample had this attribute over the whole sample period. More seriously, this variable provides no information on the market structure of generation or wholesale electricity purchasing. This variable is, however, unfortunately, the only consistently available competition variable for developing countries over the period For a fuller description of the data and a range of descriptive statistics, see Cubbin and Stern (2004), Section 4, p.19. Uruguay was a partial exception introducing licence fee funding, three years before its law came into force. We are grateful to Professor Henisz for permission to use these data. 10

11 The other main data source was World Bank data including (a) the World Bank Development Indicators (e.g. for per capita GDP in $US1995, population, etc.) and the Kaufmann governance indicators. 11

12 4 Econometric Results In what follows, we report various results. Section 4.1 covers results from a static, long-run model and section 4.2 covers dynamic models. We discuss endogeneity and causality issues in Econometric Results for Models of Generation Capacity: Static Model We started by estimating an OLS equation as a baseline. All equations reported here are modelled using a fixed effects estimator. Moving from OLS to a fixed effects model reduced the standard error of the regression by more than one-half. Given the nature of the underlying model, we would expect a fixed effects model to be more appropriate than a random effects model. For some of the equations, we tested this assumption using the Hausman test and the random effects model was consistently rejected in favour of a fixed effects model. In Table 1 below, we report some initial results including each of the four regulatory variables separately. In this table, besides GDP, we also include debt and industry control variables. The latter had consistently low t-values and are dropped in subsequent regressions. The results in Table 1 show that the individual regulatory variables are sizeable and with high t-values, although the coefficient on civil service pay is the opposite sign to the one predicted. A sample average country is estimated to increase per capita generation capacity in the long run by 18% through enacting an electricity law. In this equation, as elsewhere, the long-run elasticity of per capita electricity generation capapcity to per capita GDP is estimated as around However, the equation clearly fits the data well and provides powerful initial support for the importance of good regulation for generation investment. The problem with the results in Table 1 is that the high level of collinearity between the regulatory variables implies that the coefficient estimates on the individual effects are likely to be upward biased when taken in isolation. This conjecture is confirmed when all four regulatory variables are included in a single regression which results in the coefficient on the electricity law variable rising to 0.27 and all the other variables becoming insignificant. Omitting the law variable led to the funding variable becoming significant but with less than a 1 per cent reduction in the standard error of the regression and similarly as further regulatory variables were omitted. These results provide strong evidence that the high level of multi-collinearity between the regulatory variables significantly affects the coefficient estimates when included in combination. The standard statistical solution to this problem is to estimate a model using principal components to help better identify the effects of the individual governance elements. We did this and the results showed that only the coefficient estimate of the first principal component (accounting for 76% of the total index 12

13 Table 1: Basic Static Generation Capacity Model Results Dependent Variable = Log(Electricity Generation capacity per capita) Electricity Law Explanatory variables Real GDP per capita (log) (8.755) Type of regulator (10.667) Funding (10.363) Staffing pay (8.825) Electricity Law (5.130) Autonomous regulator (2.343) Licence funding of regulator (3.419) Civil service pay scales nonmandatory (-4.010) Debt payments as a proportion of national income Industry value added as proportion of GDP 9.38E-14 (0.021) (-0.249) 1.74E-12 (0.378) (-0.920) 4.99E-12 (1.124) (-0.645) -1.62E-13 (-0.035) (-0.599) Fixed effects Fixed effects Fixed effects Fixed effects Estimation method Adjusted R-squared S.E. of regression F-statistic Durbin-Watson No of observations Note: t statistics in parentheses variance) was statistically significant at the 5% level, with a t-value of 3.8. It is also interesting to note that: (a) the loadings of the individual components in the first principal component) were broadly similar to one another; and (b) the loading on the electricity law element was the highest. The problem with using principal components is that the results do not necessarily have any economic rationale. Hence, our preferred solution is to assemble the four regulatory variables into a regulatory index and use that index as an explanatory variable. This procedure was used in the Gutierrez (2003) telecom regulatory study and has been used extensively in the literature on the economic impact of independent central banks. (See Geraats (2002 ) for a recent survey.) 13

14 The standard procedure, which we have adopted, is to use a simple additive index. Our index takes the values 0, 1,, 4 for each country in each year depending on whether or not the country scores 1 or 0 on each of the four regulatory variables. However, as pointed out by an anonymous referee, this procedure imposes the restriction that each of the variables included in the index has the same proportionate impact on the dependent variable. This is a strong and highly debatable assumption, but, at least our index is derived from direct observation rather than from impressionistic indicators 18. In view of this and other concerns, we report below the summary results estimating alternative current and lagged versions of various regulatory variables. The fixed effects equations in Table 2 were all estimated with per capita GDP and the HZG privatisation variables as controls. In all the regressions, the coefficient on per capita GDP was around 0.7 with a t-value of 8 or more. In the table below, we concentrate on the results concerning the alternative measures of regulatory governance. The table shows that, apart from the unlagged index, all the regulatory variables are positive and significantly different from zero at the 5% level or better. It is noticeable that the lagged variables (including the 3 year plus dummy) are all larger than the contemporaneous indicators and have higher t-values. The implication that it takes time to build up the effect on regulation is supported in the age-quadratic model where the maximum regulatory impact is estimated to be at around 14 years. From Table 2, we concentrate in our dynamic modelling on the 3-year lagged index in Column 2 and the 3-year plus regulator dummy in Column 3. The lagged regulatory law has the best overall fit but, given its collinearity with the other regulatory variables, is overall less satisfactory as a descriptor of the regulatory framework. The lagged index variable implies a maximum impact on per capita generation capacity of 16% - the same as for the 3-year regulator dummy. Note that the latter, includes Ministry regulators as well as autonomous ones, although, particularly towards the end of the period, many Ministry regulators were operating with powers and duties specified in a regulatory law. 18 Estimation with a Guttman hierarchical index produced very similar results to those using a simple additive index. 14

15 Table 2: Static Generation Capacity Model with Alternative Regulatory Variables Dependent Variable = Log(Electricity Generation capacity per capita) Regulatory Index Lagged Regulatory Index 3 Year Plus Regulator Electricity Law Lagged Electricity Law Quadratic in Age of Regulator Explanatory Variables Regulatory Index (t) (1.5) Regulatory Index (t-3) (2.3) Independent or Ministry Regulator in place 3 Years or More Electricity (2.9) Act (t) (2.6) Electricity Act (t-3) (2.8) Age of Regulator (3.6) (Age of Regulator) 2 (-2.8) Estimation method Adjusted R- squared S.E. of regression Durbin- Watson No of observations Note: t statistics in parentheses Fixed effects Fixed effects Fixed Fixed Fixed Fixed effects effects effects effects One concern about the results in this and the previous table is the low value of the Durbin-Watson statistic. In the static form of the model, we would not expect this to lead to biased coefficient estimates but it may lead to over-estimated t-statistics. As a preliminary test, we estimated the column 3 model incorporating a 1 st order autoregressive process. The coefficient on the lagged residuals was 0.79 with a t- 15

16 value of However, the estimated coefficient on the lagged regulatory index was both positive (0.02) and statistically significant at the 5% level (with a t-statistic of 2.1) and was also positive and statistically significant for per capita GDP. The estimated Durbin-Watson was The result above suggests that the presence of autocorrelation in the static model does not significantly affect either the coefficient estimates or their statistical significance. However, we explore this more fully when we explicitly consider the results from dynamic models in Section Econometric Results for Models of Generation Capacity: Static Model Privatisation and Competition Incorporating the Domah data on privatisation and competition into the model did not produce any significant effects but that data had major weaknesses. Below, we report results using the better HZG data. In Table 3 below, we report estimates of the relevant coefficients in fixed effects regressions with the 3-year plus regulatory dummy as the measure of regulation. In all cases, the estimated overall fit of the equation and the coefficients on per capita GDP and the 3-year plus regulatory dummy were within 1% of those reported in Table 2 above. Results using the regulatory index also produced very similar results. On competition, the results above and others show consistent and significant long run effects on generation capacity levels of around 10-15%. However, the competition variable as defined provides no information on the amount of private company electricity generated for sale, let alone whether it was from an independent power producer selling to a single buyer or more from liberalised wholesale markets. In consequence, the results are more likely to indicate a degree of country commitment to electricity reform rather than any economic impact of competition in generation markets per se % of observations scored 1 on this variable. On privatisation, the results were varied. Unsurprisingly, there was no evidence that minority privatisation had any significant effect on generation capacity levels. But, neither did full privatisation although that only applied to 2.3% of all observations. There was some evidence, albeit weak in terms of statistical significance, that majority privatisation had a long-run positive effect on generation capacity levels of around 8-10% We tested for interaction effects between the regulatory variables and both the privatisation and competition variables but none was significant at the 10% level or better. 19 In many countries, including the UK, the legal right for new entrants to generate for resale was the first step in electricity reform but achieved little or nothing in itself. 16

17 Table 3: Static Generation Capacity Model: Alternative Privatization and CompetitionVariables Dependent Variable = Log(Electricity Generation capacity per capita) Explanatory Variables All Privatisation and Competition Variables Minority and Majority Privatisation Variables Only 50% or More Privatisation Minority Privatisation 0.05 (0.57) 0.04 (0.49) Majority Privatisation 0.09 (1.47) 0.12 (2.13) 100% Privatisation (0.07) Majority or full privatisation 0.07 (1.24) Competition (Legal right to generate electricity for resale) 0.14 (3.05) 0.14 (3.03) 0.15 (3.41) Notes: (i) t statistics in parentheses; (ii) Other independent variables in regression were per capita GDP and existence of 3-year plus regulator 4.3 Econometric Results for Models of Generation Capacity: Static Model Country Governance Effects On country governance, we firstly included as explanatory variables the Kaufmann indexes for (i) rule of law and (ii) corruption by country in The corruption index was never statistically significant in the fixed effect regressions at the 5% level or better, either as a separate variable or when interacted with regulatory variables. Estimated coefficients on the Kaufmann rule of law index were never statistically significant in its own right but sometimes approached significance when interacted with the regulatory variables. The Kaufman rule of law index was, however, highly significant in an OLS equation and led to non-significance of the electricity regulatory variable. This last result (together with the relative constancy of the cross-country rankings of general country governance indicators over long periods) is a major reason why we believe that the 20 See Mastruzzi, Kraay and Kaufmann (2003). 17

18 estimated fixed effects may well capture a large part of the country-wide institutional differences. On this last point, we also found: (i) (ii) No statistically significant correlation between the fixed effects and the Kaufmann rule of law index; but A sizeable and statistically significant interaction term between the regulatory index and the Kaufmann rule of law index in a random effects specification (a coefficient of 0.07 with a t-value of 2.3). These results provide interesting pointers to the role of governance effects in our model but are clearly far from conclusive. As a further test, we included into the static model values of the World Bank CHECKS index, which is a time-varying index of political risk. The index counts the number of veto players in a political system, adjusting for whether these veto players are independent of each other, as determined by the level of electoral competitiveness in a system, their respective party affiliations, and the electoral rules. 21 The index yields a minimum score in the absence of an effective legislature. The index score then increases linearly with the addition of subsequent veto points 22. The index is available for all of the countries in our sample for almost all years Including this index in the regressions rather than the single year Kaufmann index is a much stronger test of whether the estimates of our electricity regulatory governance effects are biased because of the absence of explicit country governance measures. If, as appears to be the case, country governance measures vary over time, the potential impact of this is not captured either by the country specific fixed effects or by inclusion of the Kaufmann index for Including the CHECKS index in our equations confirms the robustness of our estimates reported in Table 2 but the equation performance is improved by adding the CHECKS index. The estimated coefficient on the CHECKS index is correctly signed (positive) and around with t-values of around 2.1 (ie an increase of one point on the CHECKS index increases expected per capita generation capacity in the long run by 1.5-2%). Lagging the CHECKS index variable has a very small impact on the value on the coefficients for itself or any other variable.. Considering our preferred measures of regulatory governance for electricity, the coefficient estimate on the (3-year lagged ) electricity act was slightly reduced by adding the CHECKS index (from 0.20 to 0.16) but its significance level remained high with an estimated t-value at 4.1. For the 3-year lagged regulatory index, the Beck et al (2001). For further details on the definition of the index, see Beck et al (2001). We are grateful to an anonymous referee not only for the suggestion that we include the index in our modelling but also for providing the data for us for the countries and time periods in our sample. 18

19 coefficient on the regulatory variable was reduced from to but the t-value again remained high at 3.5. In addition, the coefficients on privatisation and "competition in generation were virtually unchanged from the results reported in Table 3 above. Hence, we conclude that the impact of the electricity regulatory governance variables is genuine and not just a proxy for variations in country governance. We also tested for interaction effects between the CHECKS index and the electricity regulatory governance variables. However, the estimated coefficients were both small and not significantly different from zero (with t-values of around 0.6) Finally, we estimated an equation for per capita generation capacity including the CHECKS index and per capita GDP but omitting any electricity regulatory variable. The resulting coefficient estimate for the CHECKS variable was only slightly increased (to around 0.025) indicating that the degree of collinearity between our electricity regulatory variables and the CHECKS index was very small. These results show that both sectoral regulatory governance and country governance significantly affect the level of investment in per capita generation capacity but that the impact of the sectoral variables is rather larger. However, the effects appear to be empirically separable, at least for electricity. We return to this issue in the Conclusion. 4.4 Dynamic Models and Autocorrelation In this section we discuss the results from dynamic models, including error correction models. Given the nature of the generation investment planning and construction process, we would expect quite long lags. Our two main concerns were: (i) (ii) To establish whether or not our results, particularly on the regulatory variables represented genuine causal processes or were merely spurious regressions; and To consider autocorrelation explicitly within a dynamic modelling framework, rather than as a statistical autocorrelation correction. To test whether the estimated, long-run static fixed effects levels equations are genuine rather than spurious regressions, we can check to see whether there appears to be a plausible adjustment process. The levels equation can be written as: Y it = φ i + βg it + γr it + υ it (2) which can be estimated as: Y it = f i + bg it + c R it + u it (3) 19

20 where Y it = log(electricity generation capacity per capita) G it = log(gdp per capita) R it is a regulatory governance variable; and f i is the fixed effect for country i From (3) we can calculate the implied the steady state, equilibrium, or long term value of Y it, which can be written as: Y* it = φ i + βg it + γr it (4) We now postulate a partial adjustment error correction mechanism under which the actual value of capacity changes by a constant proportion of last year s deviation from the long term value ie Y it = Y it -Y it-1 = - λ (Y it-1 Y* it-1 ) (5) where (Y it-1 Y* it-1 ) is last year s deviation from equilibrium. Equation (5), can be estimated by taking the residuals u it from (3) and estimating Y it = - λ u it-1 + e it (6) An alternative procedure is to estimate directly a differenced version of the long-run relationship, including country-specific fixed effects Y it = - λ (Y it-1 φ i - βg it-1 - γr it-1 ) + ε it (7a) = λφ - λy it-1 +λ βg it-1 + λγr i-1t + ε it (7b) More specifically, since we are particularly interested in the size and significance of the regulatory variable, R, we can impose the estimate of β from the long term levels relationship (3) and estimate Y it = λφ λ(y it-1 - bg it-1 ) + λγr it-1 + ε it (8) Equations (6) and (8) yield alternative estimates of λ, the speed of adjustment, which can be compared. In addition, we have alternative estimates of γ, the impact of regulation: firstly, from the levels equation (2); and, secondly, from the associated differenced equation (8). The validity of this procedure depends on stationarity of the data generation process. We tested for stationarity using the Pesharan-Shin W-statistic. Applying this test to the differenced equation (8), with the regulatory index as our measure of R it, the test clearly rejects the presence of a unit root in the residuals with a t-statistic of Even in the corresponding levels equation, the Pesharan-Shin W-statistic does not appear to suggest non-stationarity in the residuals, implying that our generation capacity variable, GDP and our regulatory variables are cointegrated. Very similar results were obtained on the unit root test with alternative definitions of the regulatory variable. 20

21 The key results were: (i) The estimates of λ, the speed of adjustment, were low at 0.12 but very similar as between the levels and differenced equations and both with t- statistics of 8.9. (ii) The estimates of the impact of regulation in the differenced equation (8) were positive and significant with t-statistics of 2.0 for the 3-year plus regulator and 3.2 for the regulatory index. (iii) (iv) The estimated long-run impact on per capita electricity generating capacity in the differenced equation was 24% for the 3-year plus regulator and almost 40% for the regulatory index. There was no evidence of serial correlation in the differenced equations (DW of 1.78). The overall fit of the differenced equations was good with adjusted R 2 of around 0.15 and F statistics of 4.5 or higher 24. Further estimated versions of these equations have been produced using the 3-year lagged regulatory index but also including the HZG privatisation and competition variables. These indicate a faster overall speed of adjustment with an estimated error correction term of , but this still implies a period of over 5 years before half of the regulatory effect on capacity is manifest. The full results are set out in Table 4 below. 24 See Cubbin & Stern (2004) for the full results. 21

22 Table 4: Generation Capacity Error Correction Models Dependent Variable: Log (Per Cap Generation capacity) Log (Per Cap Generation capacity) Log (Per Cap Generation capacity) Levels 1 Differences 2 Differences 3 Explanatory variables Real GDP per capita (log) (10.92) Index of regulatory governance (0-4) (t-3) (4.454) Index of regulatory governance (t-4) (1.97) Lagged Residuals from (15.44) Error Correction Term (15.40) Majority Privatisation (3.55) Majority Privatisation (t-1) (2.95) Legal right of IPP Sales (2.50) Legal right of IPP Sales (t-1) (2.82) Estimation method Fixed Effects Fixed Effects Fixed Effects Adjusted R-squared S.E. of regression F-statistic Durbin-Watson No of observations Note: t statistics in parentheses The equations reported in Table 4 demonstrate positive and statistically significant coefficients in both level and differenced equations not just for the (3-year lagged) regulatory index but also for majority privatisation and the competition variable. The estimated long run impact on per capita generation capacity from the privatisation and competition variables as derived from the differenced equation in Column 3 is: Max Regulatory Majority Privatisation Legal Right for Index Score IPP Competition 19% 20% 14% These results are very similar to the corresponding results from the static models reported in Tables 2 and 3. They provide strong support for the hypothesis that the impact of regulation and privatisation on generation capacity in developing countries is positive and sizeable but take some years to build up. 22

23 5. Endogeneity and Causality in Generation Capacity Models Endogeneity Much of the literature on regulatory effectiveness expresses concerns over the endogeneity of: (a) countries choosing to have an independent/autonomous regulatory agency; and (b) the quality of governance of that agency 26. The concern is essentially that countries with better (unobservable) governance have better functioning regulatory agencies eg because they have socio-economic characteristics that better support the rule of law, contracts and commercialisation. The problem is that it is very difficult to find good instruments ie instruments that are both correlated with the suspected endogenous variable and uncorrelated with the error term so that they can be treated as exogenous. The alternative is to try to model explicitly the decision to adopt regulatory reform, but this is a difficult task and, as yet, the results of such modelling have interesting but not very successful 27. However, it is possible to use a rank-based instrument firstly to test for the presence of endogeneity; and, secondly, to derive an IV estimator to control for any endogeneity 28. Adopting this procedure, we find that the coefficient on the residuals of the equation with the rank-based index in the basic static equation for per capita generation capacity has a t-value of 1.7, implying that there is marginal evidence of endogeneity of the Cubbin-Stern regulatory index. However, instrumenting the Cubbin-Stern index by using its predicted value in place of the actual value produces virtually identical results an estimated coefficient of with a t-value of 4.3 in the instrumented case as opposed to an estimate of and a t-value of 4.0 in the noninstrumented case. Hence, like Edwards and Waverman (2004) and Gutierrez (2003), we find some weak evidence of endogeneity of regulatory governance quality but very little change in coefficient estimates from correcting for it. 5.2 Causality We are grateful to Richard Gilbert and Jean-Michel Glachant for helpful discussions on these issues. See, for instance, Fink et al (2002) and Gutierrez (2004). See, for example, Gual and Trillas (2002). See Edwards and Waverman (2004) who follow Evans and Kessides (1993). For further discussion of the procedure and its application in this context, see Cubbin and Stern (2004), p For the reasons stated in Sections 2 and 3, we would not wish to claim that they are applicable to countries with an excess supply of generation capacity at any time during the period after This would exclude the Central and East European countries, the CIS and almost all OECD countries. 23