Total Costs in The Brazilian Efficiency Model of Distribution System Operators: An Analysis
Sandra de Sousa Xavier*, Robinson Semolini and José Francisco Moreira Pessanha
Department of engineering, Goiano Federal Institute, Brazil
Submission: July 02, 2018; Published: August 07, 2018
*Corresponding author: Sandra de Sousa Xavier, Goiano Federal Institute, Street 88, number 310, Goiânia, Brazil, 74085-010, Tel: +55 31 9 9792 0708; Email: email@example.com
How to cite this article: Sandra de Sousa Xavier, Robinson Semolini, José Francisco Moreira Pessanha. Total Costs in The Brazilian Efficiency Model of
Distribution System Operators: An Analysis. Eng Technol Open Acc. 2018; 2(2): 555585. 10.19080/ETOAJ.2018.02.555585
This study analyses the efficiency of electricity distributors in Brazil by considering total costs. The impact of the inclusion of total costs is evaluated with four different efficiency models using Data Envelopment Analysis and Stochastic Frontier Analysis. The analyses are conducted using a sample of 60 companies over two periods of time. The years 2008 to 2010 are used to calculate the efficiency frontier, and the years 2011 to 2012 are used to validate the methodology. The results show that, on average, the total costs estimated by benchmarking methods are approximately 7% lower than those observed in 2011 and 2012, that is, utilities need to reduce their total annual costs by approximately R$40 million on average.
Keywords: Efficiency; Electricity distributors; Methodologies; Electricity sector; Competitive; Environment; Incentive regulations; Operating costs; Distribution system operators; Territorial extension; Efficiency scores; Environmental variables; Tariff reviews; Remuneration; Minor components costs
Abbrevations: DSOs: Distribution System Operators; CR: Capital Remuneration; RD: Regulatory Depreciation; MC: Minor Components Costs; AC: Additional Costs; DEA: Data Envelopment Analysis; CRS: Constant Returns to Scale; VRS: Variable Returns to Scale
Since 1990, number of infrastructure sectors around the world, including the electricity sector, have initiated long reform processes, replacing rate of return regulation with incentive regulation. Although the structures and methodologies adopted by the electricity sector have changed since the reforms, the main objective of efficiency improvement has been maintained .
Rate of return regulation, which was widely used before the reform process, had an adverse effect. Specifically, it encouraged companies to overinvest to obtain greater capital remuneration. This effect is known in the literature as the Averch-Johnson effect . In this scenario, consumers are penalized by having to pay high tariffs.
Following the reform process, incentive regulation has become popular in the electricity transmission and distribution segments because it incentivizes companies to become more efficient . Under this type of regulation, benchmarking techniques are applied to detect inefficiencies during the electricity transport process. In short, these techniques aim to compare similar companies in a competitive environment .
In Brazil, rate of return regulation is partially employed in the definition of capital costs, whereas incentive regulation is fully applied in the calculation of operating costs. However, economic
regulation best practices follow a different trend: the adoption of
incentive regulation for capital and operating costs. This practice is based on the existence of a potential trade-off between the two costs . If they partially adopt rate of return regulations for capital costs and incentive regulations for operating costs, companies will simultaneously seek to raise the former and reduce the latter .
In this context, the present study proposes the use of total costs for the efficiency analysis of Brazilian distribution system operators (DSOs) from an incentive regulation perspective.
Several studies analyzing the efficiency of Brazilian DSOs have been published, but, to the best of our knowledge, no study has evaluated the economic effect of the adoption of total costs in the efficiency model. Xavier, Lima, Lima, and Lopes  propose an alternative form of efficiency analysis for Brazilian DSOs motivated by the great territorial extension. Despite the use of total costs with physical variables as a proxy, their study does not analyses the economic impact. Costa, Lopes, and Matos  evaluate operating cost models proposed by Brazilian regulators and discuss their main inconsistencies. Corton, Zimmermann, and Phillips  investigate the effect of incentive regulation on the operating costs of Brazilian DSOs, focusing on service quality. Altoé, Júnior, Lopes, Veloso, and Saurin  analyse the relationship between technical efficiency and some financial variables related to capital management using operating costs, costs related to service quality, and non-technical losses. Gil,
Costa, Lopes, and Mayrink  examine the statistical correlation
between efficiency scores and environmental variables using
operating costs as inputs.
Despite the previous research, studies that investigate
the incentive regulation effects on the total costs of Brazilian
electricity distributors are still necessary. At the moment, this
proposal is subject to an internal study by Brazilian regulator.
However, given the global trend, a shift towards total costs will
become essential. Thus, this study provides empirical evidence
of the impact of adopting total costs on efficiency analysis by
comparing four different models.
Since 2003, DSOs have been regulated by a price cap model,
which specifies an average rate under which tariffs should
be adjusted considering inflation and productivity targets (X
factor). The electricity distribution segment has completed three
tariff reviews (2003-2006, 2007-2010, and 2011-2014) and
is completing the fourth (2015-2018). During a tariff review,
capital and operating costs are redefined.
Capital costs consist of capital remuneration (CR) and
regulatory depreciation (RD). CR is the product of the
remuneration rate and the net remuneration base, which
corresponds to recognised investments and is not depreciated.
RD is the product of the average depreciation rate and the gross
remuneration base, which corresponds to total recognised
In the fourth tariff review, the previous asset base was
maintained and updated by the inflation index. New assets were
valued according to the concept of the optimised and depreciated
replacement cost, and a utilization index was applied to all
accepted assets to reduce overinvestment.
A reference price base is used to calculate the average minor
components costs (MC) and additional costs (AC), which make
up the final fixed asset value (replacement new value-RNV),
according to Equation 1:
ME-main equipment, such as circuit breakers and current
MC-fixed components associated with a particular
constructional standard, such as control cables and insulators;
AC-setting up the good, consisting of design, management,
assembly, and freight costs.
ME is valued according to the company’s price base, whereas
MC and AD are valued according to the reference price base,
which has created an incentive mechanism within capital costs.
The reference price base is structured in a modular way such
that a module is associated with each type of ME according to the
company’s group. The regulator applies a clustering technique to
segregate 63 DSOs into five groups to take into account different
levels of investment in electricity distribution systems. Each
company has an average group cost considering differences
between the concession areas. Once the prices of the ME, MC,
and AC are known, the RNV is calculated.
The Brazilian regulator applies Data Envelopment Analysis
(DEA) as an efficiency analysis, with operating costs as an
input. The outputs are the underground network, the over
ground network, the high-voltage network, distributed energy,
the number of consumers, non-technical losses, and service
quality. The sample has 61 DSOs, with mean values for the
variables during 2011, 2012, and 2013. The analysis preserves
non-decreasing returns to scale and the input orientation. The
regulator creates a confidence interval around efficiency scores
because DEA has a deterministic aspect.
From these restrictions, an operating cost target is set to
be reached over the regulatory period. At the time of review,
the target is compared to real operating costs. The difference
between real and target costs determines a regulatory trajectory.
Part of the difference is incorporated at review time, and the
remaining portion is considered in X Factor .
Unlike in the early years of reform, when regulators were
worried about operating costs, a current emerging question is
how to ensure that utilities set efficient investment levels. Over
the years, DSOs have improved their performances in response
to incentive regulations. However, significant investment is
needed over the next few years, and this need, combined with
incentives to reduce costs, accentuates a new challenge between
efficiency and investment .
This broad view of total costs has several motivations,
including the trade-off between operating and capital costs, the
freedom of companies to choose different strategies, and the
trade-off between cost efficiency and quality.
An analysis that segregates operating, and capital costs
encourages substitution between these cost categories .
Consider a benchmarking model in which operating costs are
the only input and the distribution network is the only output.
Utilities will increase investments by focusing on maximizing
output and the return to capital, resulting in greater operational
efficiency; however, tariffs will increase.
Companies can adopt different combinations of operating
and capital costs to operate and improve their networks .
When total costs are considered, a DSO is free to choose an
optimal cost composition.
In addition, total costs play an important role in service
quality analysis. As more DSOs invest in network reliability, total
costs and quality improvement marginal costs will be higher.
Therefore, a total cost model is more appropriate to evaluate this
possible trade-off .
Finally, a total cost model is considered one of the best
regulatory practices, according to Haney and Pollitt . A similar
result is presented by Mesquita , who investigates aspects
of the efficiency analyses currently employed by European and
Latin American countries. The analysis considers ten European
countries and eight Latin American countries and finds that most
of the countries surveyed use total costs.
However, adopting total costs in efficiency models can
also mean a strong incentive to reduce capital costs and may
jeopardize long-term investments . The possible adverse
effect of discouraging investment and jeopardizing the future
performance of energy distribution networks has been pointed
out as one of the possible causes for the non-adoption of
total costs by the Brazilian regulator. However, the regulator
recognizes its use as an international trend:
‘Discussions like this point toward benchmark model based
on total cost, which has been a trend in international regulatory
experience. However, a breakthrough in this direction requires
a much deeper study and certainly a space for methodological
transition and adaptation of agents’ .
This adverse effect is not observed by Cullmann & Nieswand
 when analyzing incentive regulation effects on the
investment behavior of 109 German DSOs. The results show an
increase in investments from 2009 for both public and private
companies. The authors conclude that an analysis of investment
decisions should include all institutional aspects of incentive
From a similar perspective, Poudineh & Jamasb  explore
the determinants of the investment decisions of 129 Norwegian
DSOs in the period from 2004 to 2010. The results show that the
main factors influencing these decisions are the rate of return
under the previous period’s investment, socio-economic costs,
and the lifespan of useful assets.
Cambini, Fumagalli, & Rondi  investigate the relationship
between incentives, service quality, and the investment levels
of Italian DSOs. The results indicate a causal relationship
between incentives and investment levels, and, in the process
of performance improvement, penalties are more effective than
The most recent advances in the field of efficiency,
microeconomics, and econometrics studies are focused on
efficiency frontier analysis. Given the impossibility of observing
theoretical efficiency frontiers, efficiency is determined by
empirical boundaries, estimated by observing the minimum use
of inputs given an output level or the maximum output given an
input level. This study uses DEA and Stochastic Frontier Analysis
(SFA) in estimating the efficiency of Brazilian DSOs.
DEA is a nonparametric methodology that uses real data
to measure the relative efficiency of a DMU. It was proposed
by Charnes, Cooper & Rhodes  to address the efficiencies
of companies operating in constant returns to scale (CRS) and
further extended by Banker, Charnes & Cooper  to variable
returns to scale (VRS).
This efficiency analysis can be focused on input reduction or
output expansion. The result from an input-oriented model is the
maximum reduction possible in the inputs level for a given level
of output. With an output-oriented focus, the model seeks the
maximum output quantities that can be generated by the actual
level of inputs used by the company. The efficiency scores can
vary from 0 to 1, where 1 denotes the efficient company
The majority of the DEA models consider either CRS or VRS.
For CRS model, outputs and inputs increase (or decrease) by
the same proportion along the frontier. Where the technology
exhibits increasing, constant or decreasing returns to scale along
different segments of the frontier, the VRS model is indicated.
The CRS model assesses the overall technical and scale efficiency,
while a VRS model measures only the technical efficiency.
The efficiency score of the ith company of N companies in
CRS models takes the form specified in Equation 2, where θ
is a scalar (equal to the efficiency score) and λ is a Nx1 vector
that represents the weight of each Decision-Making Unit in
the construction of the reference company. Assuming that the
companies use E inputs and M outputs, X and Y represent ExN
input and MxN output matrices, respectively. The input and
output column vectors for the ith company are represented by
xi and yi respectively. In Equation 2, company i is compared to a
linear combination of sample companies which produce at least
as much of each output with the minimum possible amount of
inputs. The Equation 2 is solved once for each company.
For VRS models, a convexity constraint Σλ = 1 is added that
ensures that the company is compared against other companies
of a similar size.
SFA, a parametric method, was originally developed by
Aigner, Lovell, and Schmidt  and Meeusen and Broeck 
and allows the estimation of the inefficiency associated with a
production function or cost.
The stochastic frontier consists of
(i) a deterministic component,
(ii) a stochastic component representing random error in
the estimation of the frontier, and
(iii) an inefficiency component for each company. It is
calculated, in most studies, using an input-oriented Cobb-
Douglas functional form with stacked data, as in Equation 3.
The SFA model allows the error to be disaggregated into two
independent components, vit and uit, and to be uncorrelated with
the explanatory variables .
The component vit is random noise that represents deviations
of the deterministic component from the frontier due to the
non-inclusion of an explanatory variable or measurement error.
We adopt the assumption that the error vit is independent and
identically distributed and normally distributed with a zero mean
and constant variance. This error term has all the characteristics
of the error term used in the classical linear regression model.
The uit component is a positive error term that reflects
the cost inefficiency of firms. This term indicates the cost
excess relative to the stochastic frontier. When this component
is null, the firm is at the efficiency frontier. Aigner, Lovell, and
Schmidt  propose using the half-normal distribution as the
probability distribution for this term, as in Equation 4:
This model is referred to as SFA-ALS. Even today, this is the
most common specification used in SFA models found in the
literature. Subsequently, other distributions have been proposed
for the u term, the most common of which are the exponential,
normal truncated, and gamma distributions .
The choice of inputs and outputs is a crucial aspect of
benchmarking methods, especially for DEA, as the discriminatory
power of these methods decreases as the number of variables
increases . Therefore, a researcher needs to be parsimonious
in choosing variables, opting for those that best describe the
There is no consensus on the best variables to describe
the electricity distribution process. Jamasb and Pollitt 
investigate the most frequently used variables in benchmarking
studies. Among inputs, the following stand out: operating costs,
number of employees, transformer capacity, and network
extension. With regard to outputs, distributed energy and the
number of consumers are the most common choices.
This study uses monetary and physical variables that are
widely adopted in benchmarking studies as well as non-technical
losses and service quality indicators. The monetary variables
are operating and total costs. The physical variables are the
same as those adopted by the Brazilian regulator in the current
tariff cycle, namely, the underground network, the over ground
network, the high-voltage network, distributed energy, and the
number of consumers. Non-technical losses and the service
quality indicators are also the same as those adopted by the
Brazilian regulator that consider the difference between actual
and expected values .
An efficiency analysis is conducted using data from 60
Brazilian DSOs from 2008 to 2012. The dataset can be found at
the website of the Brazilian regulator (www.aneel.gov.br) and
was divided into two periods: 2008 to 2010 for the efficiency
frontier calculation and 2011 to 2012 for the model validation.
The methodology used to calculate capital costs was the
same as that used by the regulator in Technical Note 185/2014
from the Economic Regulation Superintendence . Operating
costs and outputs were the same as those from Technical Note
66/2015 from the Economic Regulation Superintendence
database . Table 1 shows sample descriptive statistics.
This data shows great variability between companies,
especially for underground networks, which are only found in
the capitals of large countries.
Four distinct models are evaluated in Table 2: three DEA
models and one SFA model. The first two models were selected
to evaluate the impact of total costs on efficiency analysis. This
choice was based on the literature review presented in Section
3. The last two models were included in the analysis to validate
the DEA results using SFA, a guideline recommended by Bogetoft
and Otto .
The proposed methodology was applied to the four models
defined in Section 5.3 using data from sixty Brazilian DSOs from
2008 to 2010. Models 1, 2, and 3 were based on DEA using an
input orientation and non-decreasing returns to scale. Model
4 applied SFA and was estimated using an input-oriented cost
function. Table 3,4 shows the estimated results.
The results indicate that DSOs have average efficiency scores
of 0.70, 0.84, 0.80, and 0.81 in Models 1, 2, 3, and 4, respectively,
which indicates room for improvement.
Model 1 considers ten utilities as efficient, including three
small and seven large companies. Two of them, Eletropaulo
and Light, are located in high consumer density areas. Others that have reached the frontier do not have such high densities,
which implies relatively efficient input management. Other
utilities have an average efficiency of 0.67. This inefficiency can
be explained by low load densities and dispersed consumers,
which make such areas expensive and challenging for energy
distribution. Three CPFL Energia DSOs are considered efficient:
Piratininga, CPFL Paulista, and RGE. These results suggest a
possible advantage associated with holding characteristics,
as Semolini  also concludes. Twenty-nine utilities have
efficiency scores under 0.67, including AME, Ene. Paraíba, Ene.
Sergipe, CEMIG, and CEEE. The first three are located in the
Brazilian north or northeast, which are characterized as less
urbanized regions with the lowest monthly income . Analysis
indicates that these companies should reduce operating costs by
55% on average.
Model 2, which considers total costs as inputs, indicates
lower efficiency levels for three DSOs (Piratininga, CPFL
Paulista, and Light). New companies are considered efficient,
such as, for example, CEB, Coelce, and Cosern. Comparatively,
these companies have partial productivities that are higher than
their segment averages, especially for total costs and the highvoltage
network ratio. Therefore, some companies’ efficiencies
decrease under Model 2, whereas those of others increase, and
the segment average efficiency rises from 0.70 to 0.84. The
efficiency scores have a correlation of 0.76 with those of Model
1. Light is located at the efficiency frontier in Model 1. However,
with total costs, the DSO receives a score of 0.90; a reduction of
10% in its efficiency. On the other hand, Cepisa achieves better
results. Under Model 1, it has an efficiency of 0.59 compared
to Celtins, Coelba, and João Cesa. Under Model 2, the company
obtains a score of 0.88, and its peers are Celtins and Coelba. This
evidence indicates that Model 1 can penalise companies that are
efficient in total costs and can favour those that are efficient in
Model 1 can distort the incentives given to companies. For
example, Coelce obtains an efficiency of 0.80 in Model 1 and
of 1.00 in Model 2. These results corroborate the existence
of a possible trade-off between operating and capital costs.
Therefore, models with total costs are more appropriate for
efficiency analysis . In fact, Model 1 does not capture the
aspect of DSOs’ total costs.
In contrast with the previous models, Model 3 considers
only seven companies to be efficient. CEB, Coelce, and Cosern
have lower scores following the changes to the model, such as
the exclusion of service quality and non-technical losses and the
aggregation of the distribution network. Some companies, such
as Coelba and RGE, remain on the frontier in all three models.
The results of Model 3 results have a 0.89 correlation with those
of Model 2. In addition, the efficiency of Light is considerably
lower in Model 3, with a value of only 0.61. The company
obtained scores of 1.00 and 0.90 in Models 1 and 2, respectively.
This change can be explained by inclusion of the non-technical
loss variable, given that difference between the expected and
real values is minimal.
Model 4 estimates efficiency using SFA and estimates the cost
function using the Cobb-Douglas functional form. An exponential
probability distribution is used to estimate the inefficiency term
of the u error. The coefficient on the logarithm of the products is
shown in Table 5.
Table 5 shows that all estimates of the product coefficients
are significant at the 5% level. The significance of the variance
parameters of the error components, σ and λ, validate the
use of the SFA stochastic model. We observe that the most
important product is the distributed energy, which has an
importance of almost 50% between the three products. The
sum of the coefficients of the three products is 1.01, indicating
the possibility of constant returns to scale. The results of the
application of this model have a 0.76 correlation with those of
Model 3, since Model 3 is constructed using the same inputs and
products as this model is.
Of the sixty DSOs, thirteen companies have efficiencies
greater than 0.95, and only two companies have efficiencies
less than 0.5. Of these two DSOs, one is João Cesa, with a score
of 0.45, but in Models 1, 2, and 3, this company is considered
a benchmark. This company has the smallest outputs in the
sample, and this fact may be distorting its efficiency.
To analyses the economic impacts of the different models, we
(1) the average segment efficiency for each model,
(2) each distributor’s score divided by the average segment
(3) the product of the previous result and the average real
total cost from 2008 to 2010, and
(4) the comparison of the previous result with the average
real total cost from 2011 to 2012. The results can be seen in
Comparing the total costs estimated by Model 2 and the real
values, we find a necessary average reduction of R$37 million,
which is approximately 7% of real total costs. A similar result
was found by Yu, Jamasb, and Pollitt , who analyse the
efficiency of twelve English DSOs from 1995 to 2003. Of the
sixty companies evaluated, thirty-three exhibit total costs that
are higher than those defined by DEA. According to Model 2,
AME needs to reduce cost by R$166 million or, in percentage
terms, 35% of its total costs. Another inefficient large company
is Ampla, which spends R$331 million more relative to others.
Other DSOs have lower real total costs; RGE is a member of this
group, with a real total cost of R$575 million versus an expected
cost of R$643 million.
Coelce also uses comparatively fewer inputs, about 12%
fewer than expected. Some companies have real and expected
values that are very close, requiring no decrease or increase.
These companies include Coelba, CPFL Paulista, and Light.
Model 3 suggests an average reduction of R$49 million, or
approximately 9% of real total costs. Giannakis et al.  make
a similar diagnosis when evaluating UK utilities between 1991
and 1999. About half of companies need to reduce their costs.
This model does not include the quality and non-technical
losses variables, as in other studies [1,14,29-33,]. AME remains
inefficient, needing to reduce costs by R$162 million, which is
R$4 million less than in Model 2. Ampla needs to reduce costs by
R$364 million. As in the previous model, some utilities prove to
be efficient, such as, for example, RGE, which spent R$100 million
less than expected. Coelce maintains its good performance in
this model, and AES Sul has an appropriate level of total costs.
Model 4 presents the lowest required cost reduction, with a
value of approximately R$34 million, or 6% of costs. This result
is to be expected since SFA considers data error. This model
does not include environmental variables since they were not
significant. These results corroborate previous work, such as
that by Yu et al. , who conclude that environmental factors
do not have significant economic or statistical impacts on the
overall performances of English DSOs. The model finds the
sharpest reductions with respect to Boa Vista (58%) and João
Cesa (51%). In the previous models, the latter is considered
efficient, with opportunities to increase total costs by 3% and
8%, respectively, in Models 2 and 3. Another utility with a similar
result is Eletropaulo, which can increase total costs by R$236
million in Model 2, can increase them by R$86 million in Model
3, and should reduce costs by R$172 million in Model 4. Elektro
moved in the opposite direction, as it is evaluated positively by
Model 4 but needs improvement in Models 2 and 3.
Finally, when analyzing the results of all models, we find that,
in average percentage terms, the total costs estimated by the
benchmarking methods are not considerably smaller than those
defined by the Brazilian regulator.
Efficiency analysis is receiving considerable attention from
regulators in the electricity sector, especially in the distribution
segment. Due to the natural monopoly characteristics of the
electricity distribution process, utilities are not subject to market
This study simulated a virtual competitive scenario among
Brazilian utilities. DEA and SFA were used for efficiency analysis.
Both methods calculate an efficiency frontier based on the
evaluated company’s inputs and outputs to evaluate the impact
of total costs.
The novelty of this study is in the use of total costs as inputs
in efficiency models, specifically in the Brazilian case. Although
total costs have already been evaluated by other studies, mainly
in European countries, they have not been applied in a country
with a considerable distribution segment growth rate, such as
Four different models were studied. Comparing Model 1
and Model 2 allowed us to evaluate the impact of total costs
on efficiency, whereas the comparison between Model 3 and
Model 4 was useful to understand the robustness of the results.
In the first comparison, 88% of utilities had a higher efficiency
score in Model 2, with a mean difference of 0.14. In the second
comparison, the efficiencies of 39 companies increased with SFA,
with a correlation between the results of 0.76.
When evaluating the impact of the use of incentive
regulations in total costs, we find that DSOs need to reduce their
costs by an average of R$ 40 million per year, which is around 7%
of total costs. This efficiency gain will affect consumers, who will
pay lower tariffs.
This study evaluated the efficiency of Brazilian DSOs using
total costs as an input; future studies could focus on superefficient