Deisi Mara Ribeiro Silva

52
1 Rectorate of Post graduation and Research Postgraduate Programme Stricto Sensu in Economics WHAT AFFECTS THE EFFICIENCY OF THE OPERATING COSTS OF ELECTRICAL ENERGY DISTRIBUTORS IN BRAZIL? AN ANALYSIS USING STOCHASTIC FRONTIER Author: Deisi Mara Ribeiro Silva Supervisor: Dr. Tito Belchior Silva Moreira Brasília - DF 2015

description

Deisi Mara Ribeiro Silva

Transcript of Deisi Mara Ribeiro Silva

  • 1

    Rectorate of Post graduation and Research

    Postgraduate Programme Stricto Sensu in Economics

    WHAT AFFECTS THE EFFICIENCY OF THE OPERATING

    COSTS OF ELECTRICAL ENERGY DISTRIBUTORS IN

    BRAZIL?

    AN ANALYSIS USING STOCHASTIC FRONTIER

    Author: Deisi Mara Ribeiro Silva

    Supervisor: Dr. Tito Belchior Silva Moreira

    Braslia - DF

    2015

  • 2

    DEISI MARA RIBEIRO SILVA

    WHAT AFFECTS THE EFFICIENCY OF THE OPERATING COSTS OF

    ELECTRICAL ENERGY DISTRIBUTORS IN BRAZIL?

    AN ANALSIS USING STOCHASTIC FRONTIER

    Dissertation submitted to the Stricto Senso

    Postgraduate Programme in Economics at the

    Universidade Catlica de Braslia, as a partial

    requisite for acquiring a Masters Degree in

    Regional Economics.

    Supervisor: Dr. Tito Belchior Silva Moreira

    Braslia

    2015

  • 3

    Dissertation by Deisi Mara Ribeiro Silva, entitled WHAT AFFECTS THE EFFICIENCY OF

    THE OPERATING COSTS OF ELECTRICAL ENERGY DISTRIBUTORS IN BRAZIL? AN

    ANALSIS USING STOCHASTIC FRONTIER, submitted as a partial requisite for acquiring a

    Masters Degree in Regional Economics in the Stricto Sensu Postgraduate Programme in

    Economics at the Universidade Catlica de Braslia, on , defended and

    approved by the undersigned examination board.

    ________________________________________________________

    Prof. Tito Belchior Silva Moreira

    Supervisor

    Economics UCB

    ________________________________________________________

    Prof. Leonardo Monasterio

    Economics UCB

    ________________________________________________________

    Prof. Ricardo Silva Azevedo Arajo

    Economics UnB

    Braslia

    2015

  • 4

    I dedicate this work to my family and my

    supervisor Tito Belchior, to the Postgraduate

    Department of Economics at the

    Universidade Catlica de Braslia and to my

    friends.

  • 5

    ACKNOWLEDGEMENTS

    I am grateful to those who, in some way, assisted me throughout this journey. First and foremost

    I thank my mother Ana Gabriela and father Hamilton Silva for their motivating words and their

    assistance whenever I needed it. I am thankful to CNPq for the financial assistance, to Prof. Tito

    Belchior, Prof. Leonardo Monastrio and to all the teachers of the Post graduation Department of

    Economics at the Universidade Catlica de Braslia.

  • 6

    Were blind to our blindness. We have very little

    idea of how little we know. Were not designed to

    know how little we know.

    (Daniel Kahneman)

  • 7

    ABSTRACT

    The Brazilian electrical sector has faced many changes over the years. There have been

    transformations especially in the activities that involve the distribution of energy. This study

    measures the efficiency of the electricity distribution sector with expectations that this may help

    to establish patterns of quality and service improvement thus helping in the regulatory process.

    The objective of this study was to use benchmarking for panel comparisons, allowing us to

    compare the performance of the distributors in Brazil in different time periods. The idea was to

    estimate stochastic cost functions for 63 Brazilian distributors of electrical energy in the period

    between 2001 and 2012 using data obtained from ANEEL. These functions were used to

    calculate the efficiency of the firms and evaluate what causes inefficiency in operating costs. The

    results show that the operating costs are highly sensitive to increases in payroll and third party

    services. These two variables positively affected the dependent variable. The study also revealed

    that quality variables such as energy not distributed and DEC (duration of interruptions per

    consumer) can positively affect the efficiency of firms, though the results obtained for the latter

    were not as expected. Other quality variables such as FEC (number of interruptions per

    consumer) and technical losses seemed to negatively affect the efficiency of firms.

    Keywords: Electrical Energy Distribution. Panel. Stochastic Frontier.

  • 8

    LIST OF TABLES

    TABLE 1: Summary of the Empirical Study of Efficiency Analysis....30

    TABLE 2: Statistical Description..34

    TABLE 3: Variable Description for the Fixed Effects Model...34

    TABLE 4: Variable Description for the Stochastic Frontier Model..35

    TABLE 5: Distributing Firms divided by Groups According to the Dimension...36

    TABLE 6: Composition of the CBO`s Ofcio 376/200938

    TABLE 7: Distributing Firms Divided by Regions...39

    TABLE 8: Results for the Stochastic Frontier Model...42

  • 9

    LIST OF GRAPHS

    GRAPH 1: Brazil DEC / FEC (2013)33

  • 10

    LIST OF FIGURES

    FIGURE 1: Process for Establishing Regulation in the Electrical Sector..17

    FIGURE 2: Process for Establishing Regulation in the Electrical Sector after changes18

    FIGURE 3: Measurement and Decomposition of Cost Efficiency.26

    FIGURE 4: Measurement and Decomposition of Cost Efficiency.26

  • 11

    LIST OF ABREVIATIONS

    ANEEL National Agency of Electrical Energy

    ABRADEE Brazilian Association of Electrical Energy Distributors

    ONS National Operator of the Electrical System

    MAE Wholesale Energy Market

    CBO Brazilian Classification of Occupations

    RESEB Restructuring Project for the Brazilian Electrical Sector

  • 12

    INDEX

    1. INTRODUCTION...13

    2. BRAZILS ELECTRICAL SECTOR AND THE RESTRUCTURING

    PROCESS.........................................................................................................................16

    2.1.Preliminaries..16

    3. METHODOLOGY..21

    3.1.Fixed Effects Model..21

    3.2.Parametric Models 21

    3.2.1. Stochastic Frontier Models....23

    3.2.2. Stochastic Cost Functions..24

    3.3. Efficiency..26

    3.4. A Guide to the Literature..29

    3.5. Data Treatment .....33

    4. RESULTS AND DISCUSSION41

    4.1. Fixed Effects Model ....41

    4.2. Stochastic Frontier Model42

    5. CONCLUSIONS45

    BIBLIOGRAPHY

    APPENDICES

  • 13

    1. INTRODUCTION

    The electricity sector in Brazil has undergone many transformations. Until the beginning of the

    90s the Brazilian electricity sector had been a regulated state sector, with enterprises working

    simultaneously in the energy generation, transmission and distribution sectors (Pires, 2008).

    In the mid 90s Brazil was faced with an increase in demand that called for immediate

    investments in the sector. Institutional and operational changes that culminated in the current

    sector model were prepared based on the political-economic consensus and the regulator state,

    whose objective was to direct development policies as well as regulate the sector. As a result,

    many companies were privatized and some state agencies were created such as the regulatory

    agency, ANEEL (ABRADEE).

    Despite the new reforms, the state was unable to obtain resources that satisfied the need for new

    investments and the need to maintain the public accounts balanced and thus could not expand

    the supply of energy. The structure used at the time became unstable leading the process to be

    restructured. A rationing scheme was created, and this made clear the great challenges the

    country was facing in this area considering that a sustainable growth crucially depends on the

    reliability and quality of the energy supply.

    The growth of the economic viability of smaller electric generators, including the renewable

    sources, has introduced a new concept of how electrical systems operate, specifically, how

    energy is distributed. The growing decentralization of energy generated, which has been

    increasing in various countries of the world, has caused the distribution networks to play a great

    part on how the system operates by balancing the intermittent effects of these generators and

    increasing the quality of energy supply (ABRADEE).

    Unlike other network systems such as sanitation systems and gas, electricity cannot be stored in

    an economically viable way, and thus the need for constant equilibrium between supply and

    demand. By this we mean that the entire system can be affected or shutdown if there are

    interruptions (ABRADEE).

    Most of the research projects have been trying to improve the generation technology,

    transmission and distribution of electricity. Without questioning the importance of these studies

  • 14

    which is are to the development of the sector, it is necessary to fully understand the economic

    conditions that prevail in this market.

    Measuring the efficiency of the electricity sector can help in the regulatory process because,

    despite focusing on the electrical energy tariffs, it also permits us to establish patterns of quality

    and service improvement (TANNURI-PIANTO; SOUZA, 2009).

    In this work we use benchmarking to make panel comparisons, where we compare the

    performance of the distributors in Brazil in different time periods. Such comparisons are of great

    importance to economists and politicians considering that the development of productivity

    constitutes an essential drive for welfare improvements (BOGETOFT; OTTO, 2011).

    There are two main methods used in modern benchmarking. One is the non-parametric Data

    Envelopment Analysis (DEA) which has its roots in mathematical programming whereas the

    other is the parametric Stochastic Frontier Analysis (SFA). The latter approach allows for the

    assumption that deviations from the frontier may reflect both the inefficiencies and noise in the

    data (BOGETOFT; OTTO, 2011).

    While the non-parametric models are by nature superior in terms of flexibility, the stochastic

    frontier models have the ability to cope with noisy data. Unlike the robust estimation method, the

    stochastic models are useful in producing results that are more sensitive to random variations in

    data. They are the most flexible models when referring to the assumptions that can be made

    about data quality.

    The objective of this work is to estimate the stochastic cost functions for the Brazilian

    distributors of electrical energy in the period between 2001 and 2012. We shall use these

    functions to calculate the efficiency of these firms and evaluate the aspects of the operational

    efficiencies.

    Firstly we intend to make a preliminary exploratory analysis to observe how our input and output

    variables affect the total operational costs. We need to ensure that the effects occur as expected.

    In order to do this, we will use the panel fixed effects model to make the analysis.

    This work is organized as follows. Section two provides a brief discussion of the restructuring

    process of Brazils electrical sector. The third section describes the methodology used and other

  • 15

    methodologies proposed. Section four presents the results and makes a brief discussion of the

    obtained results. In section five we conclude this work and suggest relevant points for the

    efficiency analysis.

  • 16

    2. BRAZILS ELECTRICAL SECTOR AND THE RESTRUCTURING PROCESS

    2.1. PRELIMINARIES

    The world economy has been going since the 1970s through a great reform and deregulation

    process, especially in the electrical sector.

    Due to its territorial extension and the predomination of hydraulic sources in the electrical energy

    generation, Brazil has always faced a great challenge in the industrialization process. This

    challenge consisted in conciliating the supply of energy with the countrys economic growth

    profile.

    Firstly, to make decisions that concern the investments required for the generation, transmission

    and distribution of energy, one must consider the size, composition and regional distribution of

    the national GDP. One should consider GDP estimations made at least five years in advance. It is

    the minimum time necessary to construct an electrical plant. Secondly, defining the tariff policy

    requires a structure for energy production costs that becomes increasingly heterogeneous

    throughout time. The reasons for this include:

    - The fact that the lifespan of the installed productive capacity could remain active for

    many decades after the amortization of investments

    - The highest operation costs of the new plants, due to technological and environmental

    reasons, and the location of the hydraulic sources.

    The fiscal crisis of the 1980s, the depletion of external financial resources and the various

    attempts by the government to use public service tariffs as the remedy to deal with the inflation

    of that time, made it difficult to use the energy policy executed in the 70s. After the inflation

    control of the Plano Real, Fernando Henrique Cardosos government initiated in 1995 a program

    of reforms that included privatizing the sector and introducing a regulation model based on the

    fact that, given the current stage of technological development, the generation and

    commercialization of energy are potentially competitive, while the transmission and distribution

    are natural monopolies (LIMA JR., 2005).

    From the 1990s until the present days, three relevant facts have taken place in the electrical

    energy market:

  • 17

    - In the second semester of 1994 there was an increase in the purchase power of the classes

    of lower income, especially in the consumption of electronics and similar. As a

    consequence there were also increases in consumption and the number of consumers of

    electrical energy.

    - The second fact was the restructuring process that began in 1995 allowing for

    privatization. This generated great expectation on the final consumers, who questioned

    more about the services and the fluctuation of tariffs that occurred after the government

    stopped controlling the companies.

    - The third fact happened in the second semester of 2001, when the electrical energy was

    rationed across the country. The well-known Apago (or blackout) changed the

    consumers consumption habits all over the country. New habits such as turning off

    equipments that werent being used, maximizing the use of natural energy and replacing

    bulbs and equipment with new models of low energy consumption technologies, were

    incorporated in order to adapt the consumption to the energy supply.

    In the beginning of the 90s there was an institutional reorganization in the sector in order to

    reduce the states presence. The restructuring and regulatory process should have gone through

    the following procedures:

    Figure 1: Process for establishing regulation in the electrical sector.

    Source; Adapted from the New Model Study of the Brazilian Electrical Sector PriceWaterHouse(2004).

    The reformulation of the electrical sector formally started in 1995 with the creation of the

    Restructuring Project of the Brazilian Electrical Sector, RESEB, by the Ministry of Mines and

    Energy. In order to put the model into practice, two fundamental actions were adopted: a change

    in the legal framework and the privatization of public enterprises of electrical energy.

    Study and

    definition

    of the

    model

    Basic

    Law

    Detailed

    Regulation

    Market

    Implementation

    Privatization

  • 18

    One of the proposals was to segregate the large vertical companies (those that simultaneously

    performed the activities of generation, transmission and distribution of electrical energy), and to

    privatize the areas of generation and commercialization of electrical energy so as to establish at

    first a competitive model of free competition as opposed to the eminent monopolistic market.

    In 1996, Law no. 9.427 was implemented. The National Agency of Electrical Energy (ANEEL)

    was created to serve as the regulating agent, the mediator and inspector of the sector and to be

    responsible for the execution of concession auctions of the generation and transmission

    enterprises.

    In this same year, after the approval of the concessions law, the Brazilian government hired the

    services of the American firm Coopers & Lybrant to help it develop a new model for the national

    electrical company. The consultant suggested that the Brazilian case follow the same

    restructuring model adopted by England and adapt it to the countrys reality.

    And thus the new model known as the Restructuring Project for the Brazilian Electrical Sector,

    the RESEB Project, was implemented on August 12th

    1996, although the privatization process

    had already begun since 1995 (PIRES, 2008).

    The regulating process in Brazil suffered a few changes as seen in the figure below:

    Figure 2 : Process for establishing regulation in the electrical sector after the changes

    Source; Adapted from the New Model Study of the Brazilian Electrical Sector PriceWaterHouse(2004).

    One of the failures of Fernando H. Cardosos government was to initiate privatization before

    establishing a new regulating process. The first companies were privatized in 1995, but the

    regulating agency, the National Agency of Electrical Energy (ANEEL), only began to operate in

    1997, and in 1998, with the implementation of the two other central institutions of the new

    model, the ONS and the Wholesale Energy Market (MAE), eighteen electrical companies had

    Basic

    Law

    Study and

    definition of

    the model

    Privatization Detailed

    Regulating

    Market

    Implementation

  • 19

    already been auctioned. On the other hand, the government began to face strong opposition as a

    result of the misconduct of the process, but especially because it wasnt trivial to defend the

    advantages of the private initiative in a sector with a history of successful state enterprises. This

    political obstacle, allied to the new government priorities after the 1998 election, resulted in the

    paralysis of the privatization program in the subsequent years (LIMA JR., 2005).

    The new regulatory process was granted the legal authority to create two new institutional agents

    and a committee in MME Ministry of Mines and Energy:

    - The EPE, Enterprise of Energy Research, is a public federal firm linked to the MME,

    constituted by Law no. 10.847 and according to this law has the goal of providing

    services on the study and research areas related to subsidizing the energy planning sector,

    such as electrical energy, oil and natural gas and its derivatives, mineral coal, renewable

    energy resources and energy efficiency, among others.

    - The CMSE, Monitoring Committee of the Electrical Sector, was authorized by Law no.

    10.848/04, and according to this law it has the principal function of monitoring and

    evaluating the continuity and safety of the energy supply throughout the country. This

    means that, among other activities, it monitors the development of generation,

    transmission, distribution, marketing, import and export of electricity, natural gas and oil

    derivatives.

    Apart from these agents, there is also the important role of the CNPE, National Council of

    Energy Policy and the System Expansion Planning Coordinator Committee, which are

    responsible for the policies and industry planning, respectively.

    According to ASMAE (1999) in order for the new implemented model and the new energy

    market to work well, it is necessary to guarantee economic efficiency of the industrial park, the

    expansion of the system by the self-sustaining industry, to ensure a system that operates on high

    levels of reliability according to the quality requisites determined by society, and to guarantee

    services that are available to all agents connected to the system without any discrimination.

    The technicians working in firms that operate within these systems face a great challenge which

    is to improve the quality of the electrical energy and its associated services performing the

  • 20

    generation, transmission and distribution services with the highest quality and reliability indexes

    and the lowest possible costs (PIRES, 2008).

  • 21

    3. METHODOLOGY

    3.1 FIXED EFFECTS MODEL

    Panel data can be of great interest especially in microeconometric applications. The motivations

    that allow for the use of panel data include the desire to explore this type of data in order to

    control unobserved time invariant heterogeneity in cross sectional models and the desire to use

    panel data to disentangle components of variance, to estimate transition probabilities among

    states and to study the dynamics of cross-sectional populations, Arellano (2003). In order to

    pursue these motivations, we can use the fixed effects model.

    Considering the following equation

    = + + +

    where is a 1 x k vector of variables that vary over individual and time, is the k x 1 vector of

    coefficients on x, is a 1 x p vector of time-invariant variables that vary only over individuals,

    is the p x 1 vector of coefficients on z, is the individual-level effect, and is the

    disturbance term. If the are correlated with the regressors, they are known as FE. If the are

    correlated with some of the regressors in a model, we estimate them by treating them as

    parameters or FE.

    The FE modestly relaxes the assumption that considers the regression function constant over

    time and space. In a one-way FE model we can allow each cross-sectional unit to have its own

    constant term while the slope estimates () are constrained across units (Baum, 2006).

    Using fixed effects models to analyze panel data permits us to explore the relationship between

    predictor and outcome variables within the distributors of electrical energy. Each distributor has

    its own individual characteristics that may or may not have the capacity to influence the

    predictor variables.

    3.2 PARAMETRIC MODELS

    Suppose there is a production function f based on a technology set T. From this we have

  • 22

    () = { | (, ) }

    where x is a n dimensional input vector and y is the m = 1 dimensional output.

    In this approach we assume a priori that the production function possesses a specific functional

    form. We assume that

    () = (; )

    where the details of this function are defined by some unknown vector of parameters .

    In the parametric approach we estimate these unknown parameters from the actual observations,

    (xk, y

    k), k = 1,,K. We assume the estimated values to be .

    Unlike DEA, the stochastic frontier model is a regression model which uses the Maximum

    Likelihood principle as the estimation principle. In other words we select the value of that

    makes the actual observations as likely as possible (BOGETOFT; OTTO, 2011).

    One more aspect needs to be specified in order to implement this idea. The data generation

    process can explain the observations which are deviated from the production function. Three

    main processes are suggested in the parametric approach. These include considering any

    deviation as noise which corresponds to an ordinary regression model, considering any deviation

    as an expression of inefficiency known as the deterministic frontier, and finally assuming that

    deviations are caused by a combination of noise (the term v) and inefficiency (the term u). The

    latter suggestion refers to the stochastic frontier approach (BOGETOFT; OTTO, 2011).

    Table 1 is a summary of the parametric approaches analyzed in Bogetoft and Otto, 2011 where

    is a noise and + is inefficiency.

  • 23

    Approach Additive Multiplicative

    Regression y = f(x; ) + v y = f(x; ) exp(v)

    Deterministic y = f(x; ) - u y = f(x; ) exp(u)

    Stochastic y = f(x; ) + v - u y = f(x; ) exp(v) exp(-u)

    Source: Bogetoft and Otto, 2011

    If we look at the additive specifications we observe that the noise term v has a more increasing or

    decreasing effect on output with respect to f(x; ). On the other hand the inefficiency term 0

    will always make the observed output smaller than f(x; ).

    As we look at the Farrell and Shephard efficiency measures, which show the multiplicative

    impact, we can see that v can make the output larger or smaller, since exp() 1 when 0

    and exp() 1 when 0.

    After estimating the parametric functional form, it is possible to measure the output efficiencies

    of each firm individually.

    3.2.1. STOCHASTIC FRONTIER MODELS

    The base SFA model looks as follows after a log transformation:

    = ( ; ) + ,

    ~ (0, 2), ~ +(0,

    2), k = 1,, K

    The v term deals with the stochastic nature of the production process and captures the possible

    measurement errors of the inputs and output while the term u refers to the firms inefficiency.

    We consider these two terms independent from each other. The term v reflects the random

    disturbances that are independently and identically distributed having a normal distribution with

    a zero mean and a variance of 2. The term u reflects the random disturbances which are

    normally distributed with a mean of 0 , then there is the possibility that is not very large, as () = 0 and 0. This

    suggests relative efficiency of firm k. If < 0, then chances are that is large, meaning that

    firm k is relatively inefficient.

    Now given the cost function

    (, ) = { |(, ) }

    shows the minimum cost of producing the output combination y when input prices are w and the

    technology set is T.

    (, ) (, )

    Therefore the observed cost will be higher than the minimum cost and the cost efficiency will be

    the minimal cost compared to the actual costs

  • 28

    = (, )

    1, (, )

    We may wish to parameterize cost efficiency using the inefficiency term u such that

    =

    where 0 when 1, for (, ) .

    A multiplicative error term v is introduced so that we obtain

    = (, )

    and therefore

    = (, )

    =

    (, )

    = (, )

    We can use this equation to estimate CE. Taking the log we obtain

    log() = log((, )) + +

    Using a Cobb-Douglas function we illustrate the method where = =1 such that

    = 011

    where = + and = 1=1 to ensure homogeneity in prices.

    In logarithmic form we have

    log = 0 + 1 log 1 + + + = 0 + 1 log 1 + + log + +

    where we have an ordinary term v, and a non-negative term u that reflects the level of

    inefficiency.

    We now have the cost efficiency measured as

  • 29

    = = 01

    This equation can be rewritten as the statistically equivalent equation

    log = 0 11 log + +

    In order to deal with the homoneneity in input prices we can use one of the input prices as a

    numeraire

    log (

    ) = 0 1 log (

    1

    ) 1 log (1

    ) +

    which is equivalent to

    log = 0 1 log 1 11 (1 1 + 1) log +

    3.4. A GUIDE TO THE LITERATURE

    The calculations of the efficiency measures of frontier production and cost functions were

    introduced by Debreu (1951) and Farrell (1957), even though there are intellectual

    antecedents, [such as Hicks (1935) suggestion that a monopolist would enjoy their position

    through the attainment of a quiet life rather than through the pursuit of economic profits, a

    conjecture formalized somewhat more by Leibenstein (1966, 1975)]. Farrell considered

    analyzing technical efficiency in terms of realized deviations from an idealized frontier

    isoquant. This approach fits into the econometric approach in which the inefficiency is

    identified with disturbances in the regression model (Greene, 2007).

  • 30

    Table 1: Study of the Empirical Studies of Efficiency Analysis

    Topics Work Title Description Author Year

    Stochastic

    Frontier Model

    and DEA

    Model

    Estimating Technical

    and Allocative

    Inefficiency Relative

    to Stochast

    Production and Cost

    Frontiers

    This paper seeks to consider the

    duality between the stochastic

    frontier production and cost

    functions assuming exact cost

    minimization (technical

    inefficiency only) and inexact

    cost minimization (technical and

    allocative inefficiency) The

    inefficiency is measured using

    data on steam electric generating

    plants

    C. A. Knox

    Lovell and

    Peter

    Schmidt

    1978

    Fronteiras de

    Eficincia

    Estocsticas para as

    Empresas

    de Distribuio de

    Energia Eltrica no

    Brasil: Uma

    Anlise de Dados de

    Painel

    This work suggests a

    methodology for the efficiency

    analysis of 22 distributors during

    the period 1993 - 2001.

    Tanuri-

    Pianto,

    Sampaio de

    Souza,

    Arcoverde

    2007

    A Model for

    Technical

    Inefficiency Effects

    in a Stochastic

    Frontier Production

    Function for Panel

    Data

    An extension of recently

    proposed models for inefficiency

    effects in stochastic frontiers for

    cross-sectional data.

    Battese and

    Coelli 1995

    Benchmarking with

    DEA, SFA, and R

    This book presents recent

    developments in effiiency

    evaluation.

    Bogetoft P.,

    Otto L 2011

  • 31

    Avaliaao de

    Eficincia de

    Distribuidoras de

    Energia Eltrica

    Atravs da Anlise

    Envoltria de Dados

    com Restries aos

    Pesos.

    The analysis is carried out by

    calculating and comparing the

    efficiencies of 22 Brazilian

    firms.

    Sollero and

    Lins 2004

    Custos Operacionais

    Eficientes das

    Distribuidoras de

    Energia Eletrica: um

    Estudo Comparativo

    dos Modelos DEA e

    SFA

    Efficiency measurements of 40

    Brazilian electricity distribution

    companies obtained using the

    data envelopment analysis

    (DEA) and stochastic frontier

    analysis (SFA)

    models

    Souza e

    Pessanha 2010

    Choice of model and

    availability of data

    for efficiency

    analysis of Dutch

    network and supply

    businesses in

    electricity sector,

    Netherlands

    Eletricity Regulatory

    Service

    This work analyses various

    models that can be used to

    determine efficiency in the

    electricity sector in Holland.

    DTE 2000

    Lovell and Schmidt (1978) attempted to demonstrate evidence on total inefficiency and its

    technical and allocative components through a straightforward extension of the analysis of ALS

    (1977) and Meeusen and Van den Broek (1977). In their work they make the behavioral

    assumption that the firms objective is to minimize the cost of producing its targeted rate of

    output, subject to a stochastic production function constraint. In the case of technical inefficiency

    the firm operates beneath its stochastic production frontier, and in case of allocative inefficiency

    it operates off its least cost production path. These authors provide empirical illustration of this

    application. They estimate stochastic frontiers for a sample of US steam-electric generating

    plants (LOVELL; SCHMIDT, 1978).

  • 32

    Battese and Coelli (2005) proposed a model for technical inefficiency effects in a stochastic

    frontier production function using panel data. They applied the model to data from 14 Indian

    paddy farmers observed over a ten-year period. They concluded that for the technical

    inefficiency effects, involving a constant term, age and schooling of the farmers and year of

    observation, can be a significant component in the stochastic frontier production function.

    Furthermore, the application illustrated that the model specification allows for the estimation of

    both technical change and time-varying technical inefficiency, given that inefficiency effects are

    stochastic and have a known distribution.

    A Study that specifically analyses efficiency in electrical energy distributors is the work done by

    Tanuri Pianto and Souza (2009). In this paper they estimated production and cost stochastic

    functions for 22 Brazilian electrical energy distributors, using the single-equation method for

    panel data. From these functions they were able to obtain estimations for the efficiency levels of

    the firms; the idea is that the results obtained could provide subsidies for the authorities

    responsible for the energy regulation sector so that the factors which have impact over the sector

    may be manipulated so as to reduce the negative impacts on the efficiency levels (TANURI

    PIANTO; SOUZA, 2009).

    Another paper by Sollero and Lins (2004) uses proportional restriction to the virtual inputs and

    outputs in the efficiency analysis of 22 concessionaires of electrical energy in Brazil. The

    application of DEA (Data Envelopment Analysis) in specific situations motivated the use of

    weight restrictions so as to limit its complete freedom to variation permitted by the original DEA

    model. Many studies with weight restrictions have been proposed since Thompson and others

    (1986).

    Souza and Pessanha (2010) present efficiency measures for 40 Brazilian distributors of electrical

    energy. The measures are obtained by models of data envelopment (DEA) and stochastic frontier

    models (SFA). These two techniques can mitigate the information asymmetry and improve the

    capacity of the regulating agent to compare the performance of the distributors.

  • 33

    3.5. DATA TREATMENT

    In this study, most of the information used to estimate the cost frontiers for the 63 distributing

    firms of Brazilian electrical energy in the period of 2001 to 2012, were obtained from the

    National Agency of Electrical Energy (ANEEL). The data collected is understood to affect the

    efficiency of the firms. Some of the variables studied include operational cost, distributed energy

    in MWh and some indicators of service quality such as technical losses, the average duration of

    supply interruptions per consumer per year in hours (DEC) and the number of supply

    interruptions per consumer per year (FEC), energy not distributed in MWh (Endist), and market

    share (Share).

    Graph 1: Brazil DEC/FEC (2013)

    Brazil Supply Continuity

    Reduction of 33% in the DEC in the period 1997 to 2013 and 52% in the FEC

    SOURCE: ANEEL

    The operational costs can be further explained by variables such as expenses with materials,

    payroll, third-party services, insurance, taxes and other, as described in Nota Tcnica n

    494/2013-SRE/ANEEL, and corresponding to the expense accounts registered in account

    number 615 of the Electrical Sector Accounts Manual with a few adjustments in payroll and

    DEC (hours)

    FEC (no. of interruptions)

  • 34

    other. For payroll and third party service expenses, the price index IPCA was used while for the

    remaining costs the IGPM was considered.

    TABLE 2 Statistical Description

    Variables Obs Mean Std. Dev. Min Max

    Total Operational Cost 732 248533.9 365400.7 870.0345 2226883

    Materials 732 12992.83 20353.76 28.63848 150791.3

    Third-party Services 732 95399.47 127015.8 109.4355 820554.9

    Insurance 732 875.9841 1384.881 0 12940.87

    Taxes 732 4004.507 10509.47 -15614.53 194784.3

    Staff 732 124896.5 210992.1 587.0554 1369676

    Market (average) 610 2298928 3476958 5805.73 2.11E+07

    Technical Losses 609 464324.6 682336.4 0 4184655

    DEC 610 16.745 14.04277 0.37 102

    TABLE 3 - Variable Description for Fixed Effects Model

    Variables Description

    CO

    Operating Cost. Refers to the data of each distributor

    concerning accounts that comprise the so-called Regulatory

    operating cost, according to the Electric Sector Accounting

    Manual - MCSE.

    Materials Operational Expense classified as material costs according

    to MCSE with historical values (R$ mil)

    Third Party

    Services Operation expense classified as third party costs according

    to MCSE in historical values (R$ mil).

    Insurance Insurance related to Distribution and Commercialization

    activities.

    Taxes Operation expense classified as tax costs according to

    MCSE in historical values (R$ mil)

    Other

    Compensation for losses and damages, personal

    consumption of energy, expenses with interns and the

    Initiation Program of Labor, Consumer Council Expenses,

    expenses with internal communication and reproduction,

    collection tax, bank fees.

    Market

    (Average)

    Energy delivered by the Distributor, in MWh, we have

    High-Tension Markets (HT), Medium Tension (MT) and

    Low Tension (LT).

  • 35

    DEC

    Equivalent duration of Interruptions per consumer,

    indicates the number of hours on average that a consumer is

    without electric energy during a period, usually on a

    monthly basis.

    Technical Losses Technical losses, mainly caused by heating of the

    conducting wires of energy (Joule Effect). Commercial

    losses are linked to the theft and fraud of energy.

    ENDIST Energy not distributed. ENDIST (DEC x market)/8760

    Source: by author using data from ANEEL

    TABLE 4- Variable Description for the Stochastic Frontier Model

    Variables Description

    logdo

    Log of Operating Cost. Refers to the data of each

    distributor concerning accounts that comprise the so-called

    Regulatory operating cost, according to the Electric Sector

    Accounting Manual - MCSE.

    logm Operating Expense classified as material costs according to

    MCSE with historical values (R$ mil)

    logst Operation expense classified as third party costs according

    to MCSE in historical vallues (R$ mil).

    logseg Insurance related to Distribution and Comercialization

    activities.

    logw Payroll - Actuarial Surpluses or Deficits and retirement

    program and or Resignation are not considered.

    logt Operation expense classified as tax costs according to

    MCSE in historical values (R$ mil)

    logz

    Compensation for losses and damages, personal

    consumption of energy, expenses with interns and the

    Initiation Program of Labor, Consumer Council Expenses,

    expenses with internal communication and reproduction,

    collection tax, bank fees.

    logy (Market (average) - Energy delivered by the Distributor, in

    MWh, we have High-Tension Markets (HT), Medium

    Tension (MT) and Low Tension (LT).

    Explanatory Variables for Technical Inefficiency

  • 36

    logdec

    Equivalent duration of Interruptions per consumer,

    indicates the number of hours on average that a consumer is

    without electric energy during a period, usually on a

    monthly.

    logptec Technical losses, mainly caused by heating of the

    conducting wires of energy (Joule Effect). Commercial

    losses are linked to the theft and fraud of energy.

    logendist Energy not distributed. ENDIST (DEC x market)/8760

    logfec Equivalent frequency of Interruptions per consumer,

    indicates how often, on average, there was no interruption

    in the consumption unit (residence, business, industry, etc.)

    logshare Market share calculated as the ratio between the amount of

    energy (in MWh) distributed by the company and the total

    amount of energy distributed.

    Source: by author using data from ANEEL

    In this study, the distributors were divided into two distinct groups, as specified in Nota Tcnica

    No. 192/2014-SRE/ANEEL, in order to improve the comparison among them. One group

    comprises of larger firms and the other of smaller firms. The smaller distributors face a reality

    that is different from the others when we consider certain aspects such as the presence of the

    subtransmission network and the fight against non-technical losses.

    GROUP 1 (Large Concessionaires) GROUP 2 (Small concessionaires)

    AES SUL COELBA BOA VISTA DME-PC

    AME COELCE CAIUA EBO

    AMPLA COPEL CEA EDEVP

    BANDEIRANTE COSERN CERR BRAGANTINA

    CEAL PIRATININGA CFLO JOAO CESA

    CEB CPFL PAULISTA CHESP URUSSANGA

    CEEE ELEKTRO JAGUARI ELETROCAR

    CELESC ELETROACRE MOCOCA SANTA MARIA

    CELG ELETROPAULO SANTA CRUZ ENF

    CELPA EMG NACIONAL FORCEL

    CELPE ENERSUL COCEL HIDROPAN

    CELTINS EPB COOPERALIANA IGUAU

    TABLE 5: Distributing Firms Divided by Groups According to their

    Dimension

  • 37

    CEMAR ESCELSA CPEE MUXFELDT

    CEMAT ESE CSPE SULGIPE

    CEMIG LIGHT DEMEI NOVA PALMA

    CEPISA RGE

    CERON Source: ANEEL

    Most of the smaller distributors receive energy at a distribution level from a supplier, thus

    eliminating part of the transport and transformation costs. We classify these costs as network

    utility expense and thus are not included in the operational costs (NOTA TCNICA No.

    192/2014-SRE/ANEEL).

    In Nota Tcnica No.192/2014-SRE/ANEEL the distributors are divided into these two groups

    using algorithm clusterization to define the clusters.

    Another important aspect to be considered is with respect to the regional differences determined

    by the labor costs. The average salary can vary to a great extent among the Brazilian regions, and

    consequently, those concessionaires operating in regions where the labor cost is higher will have

    to adjust their costs to this reality. Without considering this effect we can obtain an incorrect

    efficiency measure, favoring the concessionaires located in regions where the labor cost is lower.

    For this reason, a salary variable was created and used in the efficiency calculations, adjusting

    the regions to the differences according to RAIS (Source for the construction methodology of the

    indicator). This indicator measures the position of the formed groups of concession areas in

    relation to the average (NOTA TCNICA No.192/2014-SRE/ANEEL).

    The concessionaires were grouped into eight salary groups with the three capitals, So Paulo, Rio

    de Janeiro and Braslia being treated separately. In order to distinguish the regions we created

    dummies for each region as well as dummies that define the two groups according o their

    dimension.

    It is important to state that this indicator measures the salary differences of typical occupations in

    an electricity distributor, such as the maintenance electricians. Here we compare the salaries in

    different concession areas where the employees have the same occupation.

  • 38

    The average salary was obtained by the representation of the total number of workers according

    to CBO (Brazilian Classification of Occupations), weighed by the median salary (NOTA

    TCNICA No.192/2014-SRE/ANEEL).

    =

    =1

    =1

    Where: n is the total number of CBOs considered, according to the table below.

    TABLE 6 - Composio dos CBOs do Ofcio 376/2009- SRE/ANEEL

    Descricao Resumida CBO Proprios Terceirizados Total

    Eletricista 951105 2226 24428 26654

    Eletricista de Alta-tensao 732120 8129 7193 15322

    Agente administrative 411010 6034 3755 9789

    Auxiliar tcnico de eletricidade de linhas de transmisso 732105 3957 4505 8462

    Anotador de consumo de energia eltrica, gua e gs 519940 972 5551 6523

    Auxiliar administrativo de pessoal 411005 3747 1883 5630

    Auxiliar de eletrotcnico 313105 3922 827 4749

    Atendente central telemarketing 422315 1465 2188 3653

    Ajudante de eletricista 715615 671 1586 2257

    Tcnico de eletricidade 313130 1718 459 2177

    Engenheiro eletricista 214305 1695 93 1788

    Operador de teleatendimento hbrido (telemarketing) 422310 25 1451 1476

    Contramestre (produo de energia eltrica, gs e captao de gua) 860115 1031 110 1141

    Analista de comercializao 253120 932 138 1070

    Eletricista instalador de alta e baixa tenso 731125 7 964 971

    Administrador 252105 901 37 938

    Chofer 782305 44 886 930

    Zelador 514120 13 877 890

    Eletrotcnico (produo de energia) 313110 836 51 887

    Agente de segurana ferroviria 517330 22 827 849

    Advogado 241005 296 509 805

    Operador de eclusa 861205 804 0 804

    Analista de comrcio eletrnico (e-commerce) 212405 371 407 778

  • 39

    Operador de quadro de alimentao (subestao de distribuio

    de energia eltrica) 861110 629 72 701

    Montador 374420 0 697 697

    Tcnico de manuteno eltrica 313120 585 21 606

    OUTROS - 9797 6573 16370

    Source: ANEEL

    So the concessionaires operating in regions where labor work is more expensive, will have their

    operational cost reduced for the comparison while those operating in regions where labor is less

    expensive will have their operational cost increased.

    Table 7 Distributing Firms Divided by Regions

    REGION FIRMS

    RIO DE JANEIRO LIGHT

    SOUTHEAST

    ESCELSA

    ELFMS

    ENF

    AMPLA

    EDEVP

    CPFL PAULISTA

    CNEE

    CAIUA

    BANDEIRANTE

    CPFL PIRATININGA

    ELEKTRO

    EEB

    CSPEE

    CLFM

    SAO PAULO ELETROPAULO

    NORTHEAST

    SULGIPE

    ESE

    CEAL

    CELPE

    CEPISA

    CEMAR

    COELBA

    EPB

    COSERN

    EBO

    NORTH CELPA

  • 40

    ELETROACRE

    CERON

    CERR

    AMAZONAS

    BOA VISTA

    CENTRAL-WEST

    CHESP

    CELTINS

    CELG

    ENERSUL

    CEMAT

    DISTRITO FEDERAL CEB

    SOUTH

    NOVA PALMA

    AES SUL

    CEEE

    IGUACU

    EFLUL

    COOPERALIANCA

    CELESC

    RGE

    MUXFELDT

    HIDROPAN

    ELETROCAR

    DEMEI

    FORCEL

    COPEL

    COCEL

    CFLO

    Source: ANEEL

  • 41

    4. RESULTS AND DISCUSSION

    4.1. FIXED EFFECTS MODEL

    As mentioned earlier, the idea is to first carry out a preliminary explanatory analysis permitting

    us to simply observe the effects and significance of our variables. We use the fixed effects model

    to perform this analysis.

    Eleven models are presented in tables 1 and 2 of Appendix C. The first model includes all the 63

    distributors considered for analysis in this work. The eleventh model only presents the groups

    into which the distributors were divided considering their dimension. Group one specifies the

    larger firms. The remaining models represent the different regions into which the concessionaires

    were separated according to their salary levels.

    Despite having created all these various models for analysis, we shall observe that, in terms of

    significance and the effects these variables have over the total operational cost, these models

    have produced very similar results.

    The variable materials is shown to have a 1% significance level in all the models, producing an

    increasing effect on the total operational costs as it is increased by one unit. The variables third-

    party services, insurance, taxes and staff also seem to have an increasing effect on the dependent

    variable with a 1% significance in all the models as they increase.

    The variables mentioned above make up our inputs. Now considering the output variables in this

    study we can observe for the distributed energy (Market average), in MWh, that it indicates a

    positive relationship with the independent variable although it is not significant. This, however,

    is only observed in the first model. The remaining models consider this variable to be at a 1%

    significance level elevating the value of the total operational costs as it increases.

    Two quality variables were also included in our model, technical losses and DEC (Equivalent

    Duration of Interruption per Consumption Unit). The latter indicates the number of hours on

    average that a consumer remains without current during a period (usually a monthly period).

    These two variables show a 1% significance level with an increasing effect over the total costs as

    they increase.

  • 42

    The two variables, insurance and DEC, seem to have the largest increasing effect over total cost

    while the volume of distributed energy (Market average) and technical losses seem to have

    produced very small effects on our dependent variable.

    The tables also show the intraclass correlation, rho, which indicates the percentage of the

    variance which is due to differences across panels.

    4.2.STOCHASTIC FRONTIER MODEL

    Here we present the results obtained using the Stochastic Frontier Model. We used the following

    models:

    = 0 + 1 + 2 + 3 + 4 + 5 + 6

    + 7 + 12 + 23 + 34

    + 45+ 56 + 67 + 78 + 89 +

    and

    = 0 + 1 + 2 + 3 + 4 + 5

    + 6_1

    On table 8 we can see the results obtained.

    TABLE 8 - Results for the Stochastic Frontier Estimations Using the Half-

    Normal Model

    Dependent Variable = Logdo (Operating Costs)

    Variables Coefficients Std. Error z Statistic p-value

    Logy 0.055 0.007 7.90 0.000

    Logm 0.060 0.005 12.49 0.000

    Logz 0.014 0.002 6.31 0.000

    Logw 0.508 0.006 86.04 0.000

    Logseg 0.010 0.002 4.03 0.000

    Logst 0.337 0.008 40.27 0.000

    Logt 0.012 0.003 6.94 0.000

    dr2 0.004 0.005 0.85 0.397

    dr3 -0.068 0.015 -4.47 0.000

    dr4 0.201 0.005 42.87 0.000

    dr5 0.050 0.009 5.40 0.000

  • 43

    dr6 0.119 0.010 12.12 0.000

    dr7 -0.024 0.012 -2.01 0.044

    dr8 -0.0315 0.004 -7.14 0.000

    dr9 0.028 0.010 2.90 0.004

    Constant 0.876 0.030 28.81 0.000

    Explanatory variables for technical inefficiency

    logfec -1.206 0.277 -4.36 0.000

    logendist 0.729 0.225 3.23 0.001

    logdec 1.350 0.270 5.00 0.000

    logptec -1.539 0.287 -5.37 0.000

    logshare -0.023 0.342 -0.07 0.947

    grupo_dg1 1.254 0.295 4.25 0.000

    const -0.511 4.72 -0.11 0.914

    Source: by the author

    The adjustments made to the econometric model were very good, with a correlation coefficient

    of 99% between the observed value and the predicted values as presented in Appendix B.

    As can be observed on the results table above, the estimated coefficients are all significant. The

    elasticity results show the dependent variable is highly responsive to payroll and third party

    services. The elasticity values are 0.51 and 0.34 respectively.

    The sum of the elasticities is less than 1 thus indicating decreasing returns to scale.

    It is demonstrated that the dummies for the regions were statistically significant with the

    exception of dummy dr2 which presented a coefficient of 0.004 indicating weak evidence against

    the null hypothesis.

    These results differ from those obtained by Tanuri-Pianto and Sousa (2009) in that, in their work,

    the location of the firm did not affect the costs since none of the coefficients associated with the

    regional dummies were significant. It is relevant to mention that they only divided the firms into

    four regions: Northeast, South, Southeast and Center-West. Only the Center-West region

    appeared to be significantly more efficient than Southeast, the control region.

    The explanatory variables for technical inefficiency indicate that the variable logfec (frequency

    of interruptions per consumer) has significant effects in the efficiency of these firms with an

  • 44

    estimated coefficient of -1.21. This indicates that the number of interruptions per consumer

    negatively affects the efficiency of firms. The variable logendist ( energy not distributed in

    MWh) positively affects the efficiency of the firms with a coefficient of 0.729.

    The variable logdec (duration of interruptions per consumer in hours) appeared to be statistically

    significant, positively affecting the efficiency of the firms. The results obtained can be compared

    to those obtained by Tanuri Pianto and Souza (2009). The variables logfec and logdec have the

    same effects on the efficiency of firms. The variable logfec produces the expected results which

    state that the number of interruptions contributes to increase the inefficiency. On the other hand,

    the variable logdec contradicts our expectations that the longer the interruptions the greater the

    inefficiency. Note, however that, the fact that the variable logdec includes both interruptions not

    programmed and programmed interruptions which are those destined to the network

    maintenance, can be explaining the sign of this coefficient. This is because it is reasonable to

    assume that programmed interruptions that contribute to the efficiency of the system, in general,

    have a superior duration to the interruptions which are not programmed; they would be

    predominating over the more frequent interruptions explaining thus the positive sign associated

    to the variable logdec (TANURI PIANTO; SOUZA, 2009).

    Technical losses (logptec) is seen to positively affect the efficiency of firms having a coefficient

    of -1.54.

    The market share appears to be insignificant to explain the efficiency of firms with a coefficient

    of -0.02.

    The variable grupo_dg1 (group of larger concessionaires) is also statistically significant with a

    coefficient of 1.25. This shows that it negatively affects the efficiency of firms.

  • 45

    5. CONCLUSIONS

    This work is crucial when it comes to making an assessment of the performance of electricity

    distributors in Brazil. The methodology used is greatly supported by economic theory giving us

    the tools and flexibility to use our assumptions in order to carry out systematic analyses that can

    be used for regulatory purposes.

    In this study we used the fixed effects models to analyze panel data. The choice of panel data

    allowed us to use more observations than we would in a cross-sectional data set permitting us to

    obtain more efficient estimators of the unknown parameters. According to Coelli et all (2005),

    using panel data makes it possible to relax some assumptions that are necessary to disentangle

    the separate effects of inefficiency and noise, we can obtain more consistent predictions of

    technical efficiencies and we can explore changes in technical efficiencies over time.

    The results obtained using the fixed effects estimation method allowed us to explore the effects

    of our variables over the total operational cost.

    We also estimated stochastic cost functions for the distributors of electrical energy in the same

    period. These functions were used for the calculation of the efficiency of the firms and for the

    analysis of certain characteristics of the operational efficiencies.

    The results revealed that operating costs are highly responsive to the variables payroll and third

    party services. These demonstrated a positive relationship with the dependent variable. The

    explanatory variables for technical inefficiency demonstrated curious results, some of which

    were unexpected. The quality variables such as energy not distributed and DEC (the duration of

    interruptions per consumer in hours) can negatively affect the efficiency of the firms. Other

    studies also obtained the same result for DEC which also contradict the expectations that the

    longer the interruptions the greater the inefficiency. Tanuri Pianto and Souza (2009) explain that

    DEC includes the interruptions which are not programmed as well as the programmed ones

    which are destined to network maintenance. It can be assumed that programmed interruptions

    that contribute to the efficiency of the system, usually have a superior duration to the unexpected

    interruptions.

  • 46

    BIBLIOGRAPHY

    AIGNER, D.J .; LOVELL, C. A. K.; SCHMIDT, P. Formulation and estimation of stochastic

    frontier production function models. Journal of Econometrics. North-Holland, v. 6, p. 21-37,

    1977.

    ARELLANO, M.. Panel data econometrics Oxford: Oxford University Press, 2003

    ARCOVERDE, F. D.; TANNURI-PIANTO, M. E.; SOUSA, M. C. S. Mensurao das

    eficincias das distribuidoras do setor energtico brasileiro usando fronteiras estocsticas. In:

    ENCONTRO NACIONAL DE ECONOMIA, 33., 2005, Natal. Anais

    BATTESE , G. E.; COELLI, T. J. A model for technical inefficiency effects in a stochastic

    frontier production function for panel data. Empirical Economics, v. 20, p. 325-332, 1995.

    BAUM F. C.. An Introduction to Modern Econometrics Using Stata. Stata Corp LP College

    Station, Texas, 2006

    BOGETOFT P., Otto L. Benchmarking with DEA, SFA, and R., 2011

    CATAPAN, E.A. A privatizao do setor eltrico brasileiro: os reflexos na rentabilidade e

    solvncia das empresas distribuidoras de energia. Tese (Doutorado em Engenharia da

    Produo) Programa de Ps Graduao em Engenharia da Produo, Universidade Federal de Santa Catarina, Florianpolis. 2005.

    CATAPAN, E.A. Empresas congeners do setor eltrico indicadores economic-financeiros 2001 a 2005. COPEL, Curitiba, 2006

    CHARNES, A.; COOPER, W. W.; RHODES, E. Measuring the efficiency of decision making

    units. European Journal of Operational Research, v. 2, 1978.

    COELLI, T. J.; RAO , D. S. PRASADA; BATTESE , G. E. An Introdution to Efficiency and

    Productivity Analysis. 3 ed. London: Kluwer Academic Publishers, 1998.

    COELLI, T. J. et al. An introduction to efficiency and productivity analysis. 2nd ed. Springer,

    2005.

    DTE(2000), - Choice of model and availability of data for efficiency analysis of Dutch network

    and supply businesses in electricity sector, Netherlands Eletricity Regulatory Service.

    FARRELL, M.J. The measurement of productive efficiency. Journal of The Royal Statistical

    Society, Series A, CXX, Part 3, 253-290. 1957

    FERREIRA, C.K.L. Privatizao do setor eltrico no Brasil: In: PINHEIRO, A.C. e

    FUKASAKU, K. (Orgs.). A privatizao no Brasil: o caso dos servios de utilidade pblica.

    Rio de Janeiro: BNDES-OCDE, 2000.

  • 47

    FERREIRA, C.K.L. Privatizao do Setor Eltrico no Brasil s.1, s.ed., 1999.

    FITTIPALDI, E.H.D. Leiles de comercializaode energia eltrica: Um modelo para o

    Mercado regulado no Brasil, Tese de Doutorado em Engenharia da Produo Programa de Ps Graduao em Engenharia da Produo, Universidade Federal de Pernambuco, 2005

    GUJARATI, D.M.. Basics Econometrics, McGraw-Hill., 1995

    GREENE, W. H. Econometric Analysis. 5 ed. New Jersey: Prentice Hall, 2002 (Chapter 16,

    p.429: Estimation Frameworks in Econometrics; Chapter 17, p.501-505: Maximum Likelihood

    Estimation)

    GREENE, W. H. Econometric Analysis. 5 ed. New Jersey: Prentice Hall, 2002

    JONDROW, J.; LOVELL, C. A. K.; MATE rov, I. S.; SCHMIDT, P. On the estimation of

    technical inefficiency in the stochastic frontier production function model. Journal of

    Econometrics. North-Holland, v. 19, p. 233-238, 1982.

    KUMBHAKAR, S. C.; LOVELL, C. A. K. Stochastic frontier analysis. Cambridge, 2000.

    LIMA, J.L. Polticas de governo e desenvolvimento do setor de energia eltrica. Centro de

    Memria da Eletricidade no Brasil. Rio de Janeiro, 1995.

    SCHMIDT, P.; LOVELL, C. A. K. Estimating technical and allocative inefficiency relative to

    stochastic production and cost frontiers. Journal of Econometrics. North-Holland, v. 9, p. 343-

    366, 1979.

    SOUZA, M.V.P.; SOUZA, R.C.; PESSANHA, J.F.M. Custos Operacionais Eficientes das

    Distribuidoras de Energia Eltrica: um estudo comparativo dos modelos DEA e SFA. Gesto da

    Produo, So Carlos, v.17, n.4, p. 653-667, 2010.

    SOLLERO, M. K. V.; LINS, M. P. E. Avaliao de eficincia de distribuidoras de energia

    eltrica atravs da anlise envoltria de dados com restries aos pesos. In: SIMPSIO

    BRASILEIRO DE PESQUISA OPERACIONAL, 36., 2004, So Joo del Rei. Anais

    TANURI-PIANTO, M.E.; SAMPAIO DE SOUZA, M.C.; ARCOVERDE, F.D. Fronteiras

    Estocsticas para as Empresas de Distribuio de Energia Eltrica no Brasil: Uma Anlise de

    Dados de Painel. Estudos Econmicos, So Paulo, v.39, N.1, p 221-247, Jan-Mar 2009

    VINHAES, E.A.S. A reestruturao da indstria de energia eltrica brasileira: uma

    avaliao da possibilidade de competio atravs da teoria de mercados contestveis. Dissertao (Mestrado em Economia) Universidade de Santa Catarina, Florianpolis, 1999.

    WOOLDRIDGE M. J. Introductory Econometrics: A Modern Approach. South-Western. Ohio.

    2nd

    ed., 2003

  • 48

    OTHER SOURCES

    ANEEL Agncia Nacional de Energia Eltrica. Available at: .

    ABRADEE Associao Brasileira de DIstribuidores de Energia Eletrica. Available at:

    http://www.abradee.com.br/

  • 49

    APPENDICES

    APPENDIX A

    Model with panel data Stochastic Cost Frontier Estimation

    . frontier logdo logy logm logz logw logseg logst logt dr2 dr3 dr4 dr5 dr6 dr7 dr8 dr9, uhet(logfec

    logendist logdec logptec logsh

    > are grupo_dg1) cost

  • 50

    APPENDIX B

    Postestimation predictions and partial correlation

    APPENDIX C

    Fixed Effects Model - Results

    TABLE 1

    DR1 DR2 DR3 DR4

    (1) (2) (3) (4) (5)

    Estimation

    Method

    Panel

    Fixed

    Effects

    Panel

    Fixed

    Effects

    Panel

    Fixed

    Effects

    Panel

    Fixed

    Effects

    Panel

    Fixed

    Effects

    Materials 1.003 1.059 1.090 1.127 1.080

  • 51

    .0467* .037* .039* .041* .039*

    Third-party

    Services 0.982 0.963 0.953 0.959 0.946

    .012* .007* .007* .007* .007*

    Insurance 2.909 4.708 3.947 3.670 3.866

    .490* .387* .395* .413* .394*

    Taxes 0.882 1.028 1.069 0.997 1.064

    .042* .035* .037* .044* .036*

    Payroll 1.021 0.969 0.976 0.977 0.975

    .008* .005* .005* .005* .005*

    Market

    (average) 0.000 0.006 0.006 0.005 0.005

    0.001 .000* .000* .000* .000*

    Technical Losses 0.006 0.009 0.005 0.009 0.006

    .002* .001* .001* .002* .001*

    DEC 173.538 137.088 184.135 158.660 212.574

    55.814* 41.192* 42.940* 43.761* 43.595*

    Constant -1401.738 -2096.857 -1701.212 -1283.188 -1787.610

    2302.754** 524.756* 548.877 560.353** 546.729*

    Observation

    Number 607.000 607.000 607.000 607.000 607.000

    rho 0.738 0.830 0.047 0.649 0.114

    Source: by the author. Notes: Standard Error in red. * 1% significance **

    5%significance

    TABLE 2

    Dependent Variable = Total Operational Cost

    DR5 DR6 DR7 DR8 DR9 DG1

    (6) (7) (8) (9) (10) (11)

    Estimation

    Method

    Panel

    Fixed

    Effects

    Panel

    Fixed

    Effects

    Panel

    Fixed

    Effects

    Panel

    Fixed

    Effects

    Panel

    Fixed

    Effects

    Panel

    Fixed

    Effects

    Materials 1.094 1.094 1.095 1.099 1.089 1.095

    .039* .039* .039* .039* .039* .039*

    Third-party

    Services 0.953 0.947 0.955 0.954 0.953 0.953

    .007* .007* .007* .007* .007* .007*

    Insurance 4.026 3.895 3.984 3.924 3.970 3.976

    .393* .395* .398 * .401* .396* .396*

  • 52

    Taxes 1.051 1.057 1.055 1.057 1.056 1.074

    .036* .036* .037* .037* .037* .037*

    Payroll 0.976 0.975 0.975 0.975 0.973 0.973

    .039* .005* .005* .005* .005* .005*

    Market

    (average) 0.006 0.006 0.006 0.006 0.006 0.006

    .000* .000* .000* .000* .000* .000*

    Technical

    Losses 0.006 0.006 0.006 0.006 0.006 0.006

    .001* .001* .001* .001* .001* .001*

    DEC 207.530 199.614 177.256 180.980 181.632 154.670

    43.422* 43.420* 43.207* 43.189* 43.056 * 44.381*

    Constant -391.582 -1303.512 -1583.527 -1518.414 -1600.668 -596.925

    638.194 554.403** 549.746* 556.153 548.724* 703.060

    Observation

    Number 607.000 607.000 607.000 607.000 607.000 607.000

    rho 0.214 0.102 0.034 0.003 0.114 0.038

    Source: by the author. Notes: Standard Error in red. * 1% significance ** 5%significance