Performance Measurement in the Australian Water Supply Industry
Tim Coelli and Shannon Walding
Centre for Efficiency and Productivity Analysis – School of Economics-University of Queensland – Brisbane, Australia
Centre for Efficiency and Productivity Analysis
Working Paper Series
Date: June 2005
School of Economics
University of Queensland
St. Lucia, Qld. 4072
Performance Measurement in the Australian Water
Tim Coelli and Shannon Walding
Performance Measurement in the Australian Water
Tim Coelli and Shannon Walding*
Centre for Efficiency and Productivity Analysis
School of Economics
University of Queensland
Draft – 17/June/05
Various government-owned businesses provide water supply services to Australian residents.
With the advent of recent competition and regulatory reforms in infrastructure industries in
Australia, more and more of these businesses are now facing new types of incentive-based
regulatory regimes. This has led to a desire for more information on the performance of these
businesses, both relative to each other and over time. In this study we use panel data on the
18 largest Australian water services businesses, observed over an eight-year period from
1995/6 to 2002/3, to measure the relative efficiency and productivity growth of these
businesses. Data envelopment analysis (DEA) methods are used to obtain estimates of the
multi-input, multi-output production technology. The potential use of these performance
measures in price-cap regulation is discussed, where the effects of variable selection and data
quality upon empirical results is emphasised.
* Comments provided by seminar audiences at Deakin University and the Australian National University are
The principal aim of this paper is to conduct an analysis of the performance of the urban
water supply industry in Australia. This will involve the use of empirical techniques that can
accommodate the multi-input, multi-output nature of the industry, which will be used to
provide estimates of the relative efficiency, and historical productivity growth of each of the
main urban water supply businesses in Australia. The main motivation for the study is to
provide comprehensive performance information to help regulatory authorities set CPI-X
price paths that encourage efficient performance. However, the paper contains considerable
discussion of the data shortcomings that exist and hence the degree to which these measures
should be used in a “light-handed” manner in any regulatory deliberations. Furthermore, we
indicate that considerable work is required in improving the comparability of data, especially
in the area of capital, before these types of measures can be used in a reliable manner.
Water users in Australia can be divided into two groups: (i) agricultural and (ii)
residential and industrial. The businesses that supply water to the latter group of consumers
can also be divided into two groups: (i) businesses that primarily supply water to small
regional towns and rural communities, and (ii) larger businesses that generally supply water
to the state capital cities and larger regional cities.
The latter group of large businesses are the focus of the present study. This is for two
reasons. First, these large businesses are generally owned by state governments or territories
and their prices tend to be regulated by independent regulatory agencies, while the smaller
businesses are usually owned by local town councils, without formal independent price
regulation.1 Second, the larger businesses formed an industry association known as the
Water Supply Association of Australia (WSAA) in 1995, and have subsequently been
collecting high quality data for benchmarking purposes, which they make public in an annual
publication known as WSAAfacts (see WSAA 2003). The members of WSAA are a
significant part of the Australian water supply sector, supplying water to roughly two-thirds
of the Australian population.2
1 Australia has three levels of government: (i) a federal government; (ii) states and territories; and (iii) local
councils. There are six states (New South Wales, Victoria, Queensland, South Australia, Western Australia and
Tasmania) and two territories (Northern Territory and Australian Capital Territory).
2 From this point forward, when we refer to the water supply industry we will be specifically referring to these
A description of the history and current regulatory structure of the Australian water
supply industry is difficult to provide, because it differs from one state to another. However
the following description is applicable to the majority of businesses. First, these businesses
have generally been government owned (i.e. either by a state government or a local council)
for much of the 20th century, and still are.3 Price levels have traditionally been set by the
government. These prices have often been set so as to not cover all costs of production (ie.
have been subsidised) and have generally been in the form of a fixed charge based upon the
rated value of the property being connected. Thus, cross-subsidies have been common (King
and Maddock, 1996, p20 and 26).
During the last decade a number of reforms have been implemented in the water
sector, which mirror similar changes introduced in a number of infrastructure sectors in
Australia. The businesses have been required to be more commercial in their operations and
structure, while generally remaining in government ownership. This process has become
known as “corporatisation”. The key changes relate to: (i) introduction of a corporate
structure of management; (ii) earning revenues which are sufficient for the business to earn a
commercial rate of return on its capital investment; and (iii) an independent regulator is used
to set prices at arms length from the government owner. For further detail see King and
Maddock (1996, p21).
Each state and territory has a regulatory authority that is responsible for regulating
prices charged for water by the major urban water supply businesses, plus other
responsibilities (e.g. in the electricity and gas sectors). The different state regulators use
similar but not identical methods in regulating water prices. For example, the NSW
regulator, the Independent Pricing and Regulatory Tribunal (IPART), uses CPI-X regulation,
via what is known as the “building blocks approach”. This is a form of regulation that is a
messy blend of incentive regulation and rate of return regulation, which is used by most
regulatory authorities in Australia.
In CPI-X price regulation, the regulated business is permitted to increase its prices
over a particular period (usually five years) by the change in the consumer price index (CPI)
minus an X factor. The X factor is generally called a productivity offset, because it reflects
the degree to which the regulator believes the business can improve its productivity (i.e.
3 The one exception is in Adelaide where the assets remain government owned but the government has
contracted a private company, United Water, to manage, maintain and operate the business over a 15-year
contract period ending in 2010.
reduce its costs in real terms). However, the X factor can also incorporate other things, such
as an allowance for the extra costs associated with required improvements in quality.4
The setting of the X factor value is always the subject of considerable debate. Most
Australian regulators hire consultants to study the operations of each company and identify
possible areas for cost savings. However, this approach is not without its criticisms. First, it
is generally a fairly invasive process, because the consultants require a lot detailed
information to make their assessments. Second, there is a perception that the conclusions
made by the consultants are rather “black box” in nature because they are generally difficult
to verify in a scientific manner. Third, information asymmetries tend to ensure that the
business managers always know more about the true nature of the “efficient costs” of
production, relative to the hired consultants. Fourth, the use of a business’ own performance
record to set an X factor may create incentives for the business to not attempt to improve its
rate of productivity growth because of the danger that it will lead to a higher X factor in the
next regulatory period.
These types of issues have encouraged some regulators to consider the use of industry
benchmarks in the setting of X factors.5 This generally involves the calculation of industrylevel
measures of average annual productivity growth using historical data, and/or the
calculation of firm-level measures of relative efficiency, which are measured relative to an
estimated production frontier, using a method such as data envelopment analysis (DEA) or
stochastic frontier analysis (SFA). These methods have the advantage that they are less
invasive and provide greater incentives for efficiency improvements. However, these
methods have the disadvantage that it is often difficult to capture all aspects of a particular
businesses’ operating environment in a single production model, and hence the results of
these methods need to be used in concert with additional information.
From our search of the published literature, we were unable to identify any studies
that have applied these techniques to data on Australian water supply businesses. The best
source of relative performance information currently available is WSSA (2003), which
provides a range of partial productivity measures, such as operating costs per connection and
per unit volume of water delivered, for each of its members over a number of years.
4 For example, the UK water regulator, OFWAT, actually allowed prices to increase in real terms in its first
price determination, because it required the UK businesses to make substantial investments in new capital to
achieve newly mandated quality targets. See, Saal and Parker (2000) for discussion.
5 For example, see the electricity supply case described in Diewert and Lawrence (2004).
However, WSAA (2003) does not attempt to calculate more comprehensive productivity
measures for the industry.
Hence, as noted earlier, the main aim of this paper is to fill this gap by conducting a
detailed analysis of the performance of the urban water supply industry in Australia, using
empirical techniques that can accommodate the multi-input, multi-output nature of the
industry, which can be used provide estimates of the relative efficiency and historical
productivity growth of each of the main urban water supply businesses in Australia.
The remainder of the paper is organised into sections. In section 2 we provide a brief
description of the Australian water supply industry and the factors that are likely to contribute
to differences in production costs between businesses and over time. In section 3 we review
some recent international analyses of water supply efficiency. In section 4 we describe the
data envelopment analysis methods that are used in this paper, before presenting our
empirical results in section 5, and making some concluding comments in the final section.
2. The Australian Water Supply Industry
The production process that is used to supply water to urban areas in Australia is fairly
straight forward. One generally obtains raw water from a purpose built dam (or pumps it out
of a river or groundwater aquifer), pipes it to a treatment plant for treatment, and then pipes it
to households and businesses. However, comparisons of the relative efficiency of urban
water supply businesses in Australia is a difficult exercise, because these firms operate in a
wide variety of environments. Hence, cost comparisons are likely to be influenced by a
number of factors that are not under the control of management.
Some information on the characteristics of the 18 businesses we consider in our
empirical analysis is listed Table 1.6 As can be seen, these businesses differ in various
aspects, including the size of the business, the volumes of water delivered per customer, and
the mix of residential and non-residential customers (i.e. commercial and industrial). Some
of these and other factors that could influence the costs of production across these businesses
are now discussed.
6 Note that the city of Melbourne is serviced by one wholesale water collection and treatment business
(Melbourne Water Corporation) and three water distribution businesses (City West, South East and Yarra
Valley). Thus the data listed for Melbourne Consolidated corresponds to these four businesses combined, while
the cost data listed for the three water distribution businesses includes the costs of purchasing water from the
wholesaler. Similarly, also note that Sydney, Brisbane and Gold Coast purchase bulk water from a wholesaler
and that their costs include the costs of purchasing water from the wholesaler.
[Table 1 here]
A high percentage of non-residential customers is likely to be associated with higher
costs per connection because these customers tend to consume higher volumes, but it is also
likely to reduce costs per mega litre because of reductions in connection related costs and the
fact that some industrial customers require lower levels of water treatment. From Table 1 we
see that the percentage of non-residential customers is fairly uniform across the Australian
businesses, with an average of 8.55%, however a few businesses do deviate to some extent,
from a high of 12.24 in Goulburn Valley to a low of 4.62 in Gosford.
A high percentage of water from non-catchment sources (such as rivers and
groundwater) is likely to be associated with lower capital costs (i.e. less dams needed) but
conversely is likely to be associated with higher operating costs, due to larger amounts of
pumping and treatment required. From Table 1 we see that these Australian water businesses
derive less than 20% of their water from non-catchment sources (ie. pumping from rivers and
groundwater), on average. However, three businesses derive over half of their water from
non-catchment sources, namely Hunter, Perth and Barwon.
Higher average rainfall is likely to reduce the capital costs of water catchment
because smaller dams are required since they are replenished more quickly. Average rainfall
levels vary significantly across these businesses, from a low of 458mm in the Goulburn
Valley in the south to a high of 1,953mm in Darwin in the tropical north.
Temperature differences can have a range of effects. Higher average temperatures
can increase the demand for watering gardens and hence increase volumes per customer,
while a wide range of temperatures over the year can result in a high peak to average flow
ratio, the latter leading to larger capital costs per unit volume delivered, because the network
needs to be built to accommodate the peak. Information on average maximum temperatures
and peak to average flows are presented in Table 1, where we see that average maximum
temperatures do not vary significantly, with all but Darwin (with 33 degrees Celsius) lying in
the range from 19 to 26 degrees. The data on the ratio of peak to average flow is also fairly
uniform, with most values lying in the range from 1.5 to 2.0. This is not a wide amount of
variation, given that the numerator in this ratio depends upon water demand on a single day
in the year.
A higher network density is likely to reduce costs associated with water distribution
because less pipe infrastructure is needed per connection.7 Information on the number of
connections per km of mains pipeline is presented in Table 1. This shows a range of
densities, from around 60 to 70 for most large cities to around 30 for those businesses that
service the regional centres.
A large business size may result in lower costs because of scale economies, but if the
large size is associated with serving a large city, then this may also increase the capital costs
associated with collecting water, as discussed above. The data on number of connections in
Table 1 shows that there is substantial size variation, from around 50 thousand properties for
a number of the regional businesses to over 1.5 million in Sydney, the largest city in
A hilly topography can affect costs because of the extra pumping costs that are
generally incurred. The data on electricity usage per connection (which is highly correlated
with pumping activity) reported in Table 1, exhibits a wide range of values, from a low of 6
kw per connection for Yarra Valley, which receives all of its water supply from catchments
up in the hills outside Melbourne, to a high of 332 for Adelaide, which needs to pump over
40% of its water from the Murray River.
The soil type can be important, with clay soils contributing to more pipe breakages,
especially for the older terracotta pipe networks, and hence higher maintenance costs.
However, clay soils can also mean a better seal on the dams and hence lower water losses
contributing to lower water catchment capital costs. Information on soil type differences is
not readily available, however it is known that cities such as Perth have a low clay content in
their soils, relative to some other cities in Australia.
Differences in demand management policies (e.g. water use restrictions) can also
influence costs via its effect on volumes per customer and also upon the ratio of peak to
average flow. Once again, information on this factor is not readily available (nor easy to
define), however it is known that these businesses have placed varying degrees of emphasis
on demand management in recent years. For example, due to water catchment constraints,
Gold Coast Water has been active in this area for some years, with the results of this activity
reflected in their low peak to average ratio in Table 1.
7 This is a view that is commonly expressed by both regulators and regulated water businesses in Australia.
However it is interesting to note that Feigenbaum and Teeples (1983, p674) hold the opposite view in that they
expect the costs of US water companies to increase with density because of the need for “more hydrants, higher
water pressure and greater peak capacities for fire protection”.
Other differences in local regulations and policies, such as water pressure standards,
minimum capacity standards (set by fire authorities), water quality standards and reliability
standards, can also affect costs, however these are generally fairly uniform across Australia.
The above list of issues is not complete but does include some of the key cost drivers
in this industry. What is clear from this discussion is that comparative performance
measurement in the urban water supply industry in Australia is a challenging exercise. The
model that we use in the empirical section of this paper will not be able to accommodate all
of these factors completely, due to data constraints and degrees of freedom constraints.
Hence the performance measures obtained should clearly be used carefully.8
3. Review of literature
In this section we review some studies that have conducted economic analyses of urban water
supply businesses using empirical modelling techniques such as regression analysis, data
envelopment analyses and stochastic frontier analysis. The review does not include any
Australian studies because we were unable to find any Australian studies in the published
The question of the relative efficiency of public versus privately owned utilities led to a
number of econometric analyses of water supply utilities in the USA in the late 1970’s and
1980’s. First, Crain and Zardkoohi (1978) estimate a Cobb-Douglas cost function and
conclude that the public firms have significantly higher costs, relative to private firms. Their
model involved a (logarithmic) regression of cost on output quantity, labour price, capital
price and an ownership dummy variable. The output measure used was volume of water
delivered while the cost measure was the sum of operating, maintenance and depreciation
costs. This output measure can be criticised on the basis that it assumes a homogenous
output, while the cost measure is also sub-optimal because it excludes the opportunity cost of
A later study by Bruggink (1982) also comes to the same conclusions regarding the
superiority of private firms using a similar approach. These two studies are then criticised in
a subsequent study by Feigenbaum and Teeples (1983), who argue that the empirical work in
8 This statement will be made even more strongly in later discussion where we discuss some of the data
comparability issues, especially on the capital side of things.
these two previous studies is flawed because of: (i) the use of volume as the only output
measure; (ii) the use of a simple functional form;9 and (iii) the omission of relevant factor
prices. They go on to specify a cost function model in which output is modelled using a
hedonic function (which includes variables reflecting metering, treatment levels, density,
capacity utilisation, customer size and water purchases); a more general translog functional
form is specified; and an electricity price variable is included (in addition to labour and
capital prices). They conclude that there are no significant differences in the costs of public
and private firms. However, for some reason they exclude capital costs from their cost
measure, which seems strange given that capital costs generally exceed operating costs in
most network utilities.
Byrnes, Grosskopf and Hayes (1986) also address the private/public issue, but they
instead decide to use the linear programming technique known as data envelopment analysis
(DEA) to estimate levels of technical efficiency for each firm in the sample. They argue that
their approach should be preferred to the cost function methods because of: (i) a lack of
reliable data on (the economic notion of) capital costs; (ii) input quantity data being more
reliable than input price data; (iii) no need to impose a function form; and (iv) the method
produces estimates of best practice performance as opposed to average performance. They
specify a production model with one output variable, volume of water delivered, and seven
input variables: ground water, surface water, purchased water, part-time labour, full-time
labour, length of pipeline and storage capacity. They conclude that there are no significant
differences in the technical efficiency scores (nor the scale efficiency scores) of private
versus public firms.
On face value, the Byrnes et al (1986) study could be criticised for not including more
output indicators (as used in the Feigenbaum and Teeples study). However, as they point out,
the input variables used are likely to control for a number of these differences in output
characteristics. For example, the use of the three water source variables will ensure that firms
with similar water source mixes will be benchmarked with each other,10 while the use of two
capital proxies (storage capacity and length of pipelines) should mean that firms with similar
network densities will generally be benchmarked with each other.
9 The Cobb-Douglas functional form is restrictive in the sense that it imposes constant input elasticities and
elasticities of substitution which are equal to one (Coelli, Rao & Battese, 1998:201).
10 This is because in output orientated DEA the method measures technical efficiency as the maximum amount
by which output can be expanded, while holding the input quantities (and hence mixes) fixed.
Teeples and Glyer (1987) provide a comparison of the models of Crain and Zardkoohi
(1978), Bruggink (1982) and Feigenbaum and Teeples (1983), using data on water utilities in
California, and argue that the differing conclusions in these earlier papers can be put down to
the model restrictions implicit in the earlier papers.
Interest in the issue of public versus private ownership of water supply companies in
the US waned for a decade or so until another round of studies surfaced in the mid 1990’s
authored by Bhattarcharyya and colleagues: Bhattarcharyya, Parker and Raffie (1994) and
Bhattarcharyya et al (1995a,b). These three studies also estimated econometric cost
functions, but used more up-to-date data and looked at a number of alternative
methodological approaches, such as (i) specifying a short run cost function (with capital
quantity specified as a regressor); (ii) estimating the cost function in a system with first order
equations; (iii) estimating a shadow cost system to reflect possible deviations from cost
minimising behaviour; (iv) estimating the cost function using stochastic frontier techniques;
(v) including quality variables such as system disruptions and water losses in the model, etc.
Lambert and Dichev (1993) also conducted a comparison of privately and publicly
owned water utilities. They used data on 238 public and 32 private firms from a 1989 survey
conducted by the American Water Works Association (AWWA) and measured technical,
allocative and scale efficiency using DEA. The single output variable used was total water
delivered, while the four input variables used inputs were: annual labour use in hours; British
thermal units of energy used; value of material inputs used; and value of capital. The study
finds that technical inefficiency is the main source of inefficiency and that there are no
significant difference between private and public firms.
The 1990’s also heralded the arrival of several studies using UK data, motivated by the
privatisation moves in the early 1990’s in the UK. These include the stochastic cost frontier
analysis study by Lynk (1993); the comparison of DEA and regression methods in Cubbin
and Tzanidakis (1998); the DEA studies of Thanassoulis (2000a,b) the cost function study of
Ashton (2000); the Tornqvist total factor productivity (TFP) index study of Saal and Parker
(2000) and the stochastic cost frontier study of Saal and Parker (2001).
In one of his SFA cost function models, Lynk (1993) studied the efficiency of 22
privately-owned water companies from 1984/85 to 1987/88. The dependent variable was
operating cost, with the regressor variables being one output variable (water supplied); one
input price variable (unit labour cost), and dummy variables for time and geography. The
model was unusual in that it did not include a fixed capital variable, and did not attempt to
accommodate the effects of customer size and network density.
Cubbin and Tzanidakis (1998) used 1994/95 UK water industry data to conduct a
comparison of regression analysis (RA) and DEA. A measure of operating expenditure
adjusted for factors outside the companies’ control and unrelated to observable cost drivers
was used as the sole input variable. Outputs were water delivered, length of mains and the
proportion of water delivered to non-households. The results indicate differences in rankings,
and the authors conclude that DEA is best used when large samples are available, although
RA does not put individual weights on variables and as such may not be as fair to individual
Thanassoulis (2000a and 2000b) undertook a data envelopment analysis of water
distribution in the UK using data obtained from OFWAT. He included the input of operating
expenditure, and argued for the exclusion of capital costs from the input(s) because OFWAT
saw no convincing evidence that operating expenditure and capital expenditure were
inversely related. Output measures used include number of properties connected, length of
mains, volume of water delivered and pipe bursts. The choice of length of mains and pipe
bursts as output variables are arguably controversial. The mains variable was included to
attempt to capture the effects of network density. However, given that mains are a capital
input, the use of mains as an output variable could perhaps signal to firms that more input is
better. Mains bursts were included to attempt to reflect the fact that certain networks are
more susceptible to bursts and hence require more reactive maintenance. However, one
could alternatively argue that one would normally require a water company to attempt to
minimise pipe bursts (through better maintenance) rather than maximise them. Once again,
this output variable could send rather unusual incentive signals to the firm being assessed, in
the medium term.
In addition to these US and UK studies, a handful of additional studies have appeared in
recent years. For example, the cost function study of French water supply businesses in
Garcia and Thomas (2001); the SFA cost frontier study of water supply industries in Asian
countries by Estache and Rossi (2002) and the DEA study of Mexican water supply
businesses in Anwandter and Teofilo (2002). These papers tend to use similar methods to
those discussed above.
4. Performance measurement methods
Simple ratio measures, such as water delivered per employee and operating costs per
connection, are widely used performance measures. The popularity of these ratio measures,
which we will call “partial productivity measures”, stems from the fact that they are easy to
construct and also easy to interpret. However, in many cases these ratio measures are
unreliable indicators of the “true productivity” of the business. This is because a particular
business could have high operating costs per connection because it is poorly managed and
wasteful, or it alternatively it could be due to factors not under the immediate control of the
managers, such as (i) having high volumes per connection (due to a large proportion of nonresidential
customers or due to climatic factors); (ii) servicing an area with a low population
density; (iii) owning assets which have a high average age and hence require more
maintenance costs; (iv) being a small business and hence suffering from diseconomies of
scale; and so on.
The key problem with this ratio measure of operating costs per connection is that it is
a partial productivity measure, in that it does not include all information on the inputs and
outputs used by the firm.11 For example, it does not include output characteristics related to
volumes per connection nor network density, and it ignores capital inputs, such as pipes and
pumps. Furthermore, it does not take account of differences in the size of the business.
These problems could perhaps be addressed by dividing the sample of firms up into a number
of groups according to business size, and then according to volumes per customer, and then
according to network density, and then according to capital intensity – but soon you would
find that most cells in the four dimensional table would contain one firm or fewer, and hence
a benchmarking exercise would not be sensible.
As an alternative to this, we use a method known as data envelopment analysis (DEA)
in this study. This technique uses linear programming methods to build a piece-wise surface
over data (on input and output quantities) for a sample of firms and then assesses the
efficiency of each firm by measuring the distance that each data point lies below the best
practice production frontier. This technique has the advantage that it can accommodate
multiple inputs and multiple outputs, and produces information on “peer firms” for each of
the inefficient firms. That is, those firms that have a similar input mix, output mix and scale
of operation (to the particular inefficient firm), but are located on the frontier surface, and
11 See related discussion of the gas supply industry in Carrington, Coelli and Groom (2002).
hence are producing the same output with fewer inputs. This method will be described in
more detail shortly.
As is evident from the review of literature in the previous section, other techniques,
such as ordinary least squares (OLS) regression, stochastic frontier analysis (SFA) and total
factor productivity (TFP) indices calculated using price-based index numbers (PIN), have
also been used in analyses of water industry performance in overseas studies. OLS methods
are well known and easy to implement, however they could be criticised on the basis that
they require the specification of a functional form for the production technology and they
provide information on average performance rather than frontier performance.
SFA is an econometric technique that addresses this latter problem, by specifying a
composed error term, with one part used to capture data noise and the other inefficiency.
However SFA methods still require a functional form to be specified, plus distribution forms
for its composed error structure. PIN methods, such as the popular Tornqvist TFP index,
suffer from the problem that they require access to reliable price information (which is often
difficult to obtain) plus they do not explicitly accommodate scale effects.
The DEA method used in this study is a frontier method that does not require
specification of a functional form or a distributional form, and can accommodate scale issues.
Hence it can address the above problems. However, DEA has the disadvantage that it does
not explicitly accommodate the effects of data noise, while OLS and SFA methods do. On
balance we have decided to use DEA methods here because we believe that data noise is less
of an issue in an industry such as water supply, where accounting standards are high (relative
to the case of small farmers in a developing country where SFA would be a wiser choice),12
and because DEA is able to readily produce rich information on technical efficiency, scale
efficiency and peers. However, in future work we plan to also use SFA methods to
investigate the sensitivity of our results to the choice of methodology.13
Efficiency measurement using DEA
DEA uses linear programming (LP) to obtain the measures of technical efficiency (TE). The
input-orientated DEA LP is set up so as to maximise the TE score of the i-th firm, subject to
12 See Coelli (1995) for further discussion.
13 See Coelli, Rao & Battese (1998) for further details regarding these various methods and their relative merits.
production remaining within the feasible set of production possibilities.14 This involves the
solution of the following LP problem.
st -yi + Yλ ≥ 0,
θxi – Xλ ≥ 0,
λ ≥ 0, (1)
where yi is a M×1 vector of outputs produced by the i-th firm, xi is a K×1 vector of inputs
used by the i-th firm, Y is the M×N matrix of outputs of the N firms in the sample, X is the
K×N matrix of inputs of the N firms, λ is a N×1 vector of weights (which relate to the peer
firms) and θ is a scalar measure of TE, which takes a value between 0 and 1 inclusive. This
problem is be solved N times – once for each firm in the sample.15
The above DEA LP has become known as the constant returns to scale (CRS) DEA
model because the resulting technology will be a CRS technology. Thus, the efficiency
scores obtained from this DEA model will be influenced by scale effects, if they exist. This
may not be desirable in some cases, since firms cannot always influence scale in the short
run. The above CRS DEA LP can be adjusted in order to allow a variable returns to scale
(VRS) DEA technology. This is done by adding a convexity constraint to the original
problem, resulting in the following LP,
st -yi + Yλ ≥ 0,
θxi – Xλ ≥ 0,
λ ≥ 0, (2)
where N1 is a vector of ones. This VRS LP produces technical efficiency scores that are
either greater than or equal to those from the CRS problem. This means that the variable
returns to scale specification gives “pure” technical efficiency scores, which are free of scale
14 DEA models can be either input or output orientated. In the input orientation the efficiency scores relate to
the largest feasible proportional reduction in inputs for fixed outputs, while in the output orientation it
corresponds to the largest feasible proportional expansion in outputs for fixed inputs. It is common practice to
use an input orientation in analyses of network utilities because the firms are generally required to supply
services to a fixed geographical area, and hence the output vector is essentially fixed. For example, see
discussion in Coelli et al (2003, p41).
15 The discussion here is based on that in Coelli, Rao & Battese (1998).
A scale efficiency (SE) score can be derived (for each firm) by dividing the CRS TE
score by the VRS TE score. This SE score also takes a value between 0 and 1 inclusive.
Productivity measurement using DEA
If one has access to suitable panel data, Fare et al (1994) have shown that DEA frontier
construction methods can be used to obtain estimates of Malmquist TFP index numbers. This
approach also has an advantage relative to PIN TFP methods (e.g. Törnqvist TFP indices)
• price data are not required;
• the TFP indices obtained may be decomposed into components:
o technical change (frontier-shift),
o technical efficiency change (catch-up).
The one principal drawback of the Malmquist methods is that panel data are required,
while the PIN methods may be calculated with only a single observation in each time period.
However, this is not an issue in this study because we have panel data on 18 firms over an
The Malmquist TFP index measures the TFP change between two data points by
calculating the ratio of the distances of each data point relative to a common technology. If
the period t technology is used as the reference technology, the Malmquist (input-orientated)
TFP change index between period s (the base period) and period t is can be written as
( ) ( )
( s s)
s s t t
i d y , x
d y , x
M y , x , y , x = . (3)
Alternatively, if the period s reference technology is used it is defined as
( ) ( )
( s s)
s s t t
d y , x
d y , x
M y , x , y , x = . (4)
Note that in the above equations the notation di
s(xt, yt) represents the distance from the period
t observation to the period s technology. When t = s this distance is equivalent to the
technical efficiency scores defined earlier. A value of Mi greater than one will indicate
positive TFP growth from period s to period t while a value less than one indicates a TFP
As noted by Färe, Grosskopf and Roos (1998), these two (period s and period t)
indices are only equivalent if the technology is Hicks output neutral. That is, if the output
distance functions may be represented as di
t(x,y) = A(t)di(x,y), for all t. To avoid the
necessity to either impose this restriction or to arbitrarily choose one of the two technologies,
the Malmquist TFP index is often defined as the geometric mean of these two indices. That
( ) ( )
i s s t t d y , x
d y , x
d y , x
d y , x
M y , x , y , x
= × , (5)
An equivalent way of writing this productivity index is
( ) ( )
i s s t t d y , x
d y , x
d y , x
d y , x
d y , x
d y , x
M y , x , y , x
= × , (6)
where the ratio outside the square brackets measures the change in the input-oriented measure
of Farrell technical efficiency between periods s and t. That is, the efficiency change is
equivalent to the ratio of the Farrell technical efficiency in period t to the Farrell technical
efficiency in period s. The remaining part of the index in equation 5 is a measure of technical
change. It is the geometric mean of the shift in technology between the two periods,
evaluated at xt and also at xs. Thus the two terms in equation 6 are:
Efficiency change =
( ) s s
d y , x
d y , x
Technical change =
d y , x
d y , x
d y , x
d y , x
The four distance measures in equation 5 are calculated by solving four DEA-like linear
programming (LP) problems. The required LPs are:16
t(yt, xt) = minθ,λ θ,
st -yit + Ytλ ≥ 0,
θxit – Xtλ ≥ 0,
λ ≥ 0, (9)
16 Note that these are CRS DEA models. CRS is required to ensure that the TFP index includes scale effects.
For further discussion see Coelli et al (1998).
s(ys, xs) = minθ,λ θ,
st -yis + Ysλ ≥ 0,
θxis – Xsλ ≥ 0,
λ ≥ 0, (10)
t(ys, xs) = minθ,λ θ,
st -yis + Ytλ ≥ 0,
θxis – Xtλ ≥ 0,
λ ≥ 0, (11)
s(yt, xt) = minθ,λ θ,
st -yit + Ysλ ≥ 0,
θxit – Xsλ ≥ 0,
λ ≥ 0, (12)
These four LP’s must be solved for each firm in the sample. Thus when there are 18 firms
and two time periods, 74 LP’s must be solved. As extra time periods are added, one must
solve an extra three LP’s for each of the 18 firms (to construct a chained index for each firm).
Thus we need to solve 74+3*18*6=398 LP’s in this instance.
5. Data and empirical results
The data used in this exercise is taken from WSAAfacts (2003, 2001). The data produced in
these WSAAfacts publications is very detailed and comprehensive. However, since the data
was not collected with this study specifically in mind, we do note that some of the variables
we use are clearly sub-optimal, and hence our results should be viewed with caution and
should only be viewed as preliminary in nature.
Two data sets are used in this section. The first is annual data on the 18 firms from
the 2002/03 financial year. This is the most recent available information and hence will be
used to calculate the technical efficiency and scale efficiency scores. The second data set is
panel data, containing data on these 18 firms over an eight-year period from 1995/96 to
2002/03. When discussing this latter data set the issue of the selection of appropriate price
deflators becomes important.
The selection of the input and output variables that are to be included in a DEA model
is a complicated exercise. The decision process in this study is guided by our discussion of
the cost drivers in the industry; our review of the overseas literature; our knowledge of the
available data in WSAAfacts; and by the degrees of freedom constraints that we face when
using such a small sample size. We have decided to limit our attention to models that involve
no more than four variables, due to our degrees of freedom constraints.
We have chosen two output variables:
• number of properties connected (PROP), and
• volume of water delivered (WDEL),
and two input variables:
• operating expenditure (OPEX), and
• capital (CAP).
This set of output indicators is a set that is often used in network industries, such as water,
electricity and gas. They are included to ensure that firms with similar average customer
sizes are benchmarked with each other.17 The other main output attribute, network density, is
accommodated indirectly in this model by ensuring that the input set contains both a capital
and a non-capital variable. Given that the capital intensity of these firms is primarily
determined by their network density (ie. sparse networks tend to have higher amounts of
pipeline capital relative to OPEX because their customers are further apart) this will tend to
ensure that high density firms are benchmarked with similar firms and vice versa.
Some previous studies have broken up OPEX in to smaller variables, such as labour
and non-labour measures. This allows one to use physical measures of labour if available.
Since we had no data on the labour input, this was not a choice that was available to us.
Furthermore, given the amount of outsourcing that is prevalent in many of these firms, the
distinction between labour and non-labour OPEX would be rather arbitrary. In addition,
given our degrees of freedom constraints, the inclusion of an additional variable in the model
would not be wise in any case.
The choice of an appropriate measure of capital input is always a challenge in
empirical analyses such as this. Major water supply asset groups include pipes, pumps,
treatment plants and storage, plus other groups such as vehicles, buildings and equipment.
Given that detailed and comparable data on these various groups were not available and given
17 An alternative set of output indicators could be to have volume delivered to small customers and volume
delivered to large customers. However, such data was not available to us, and would be unlikely to provide a
better fit if available.
our degrees of freedom constraints, the obvious option was for us to find a monetary measure
that could provide a reasonable proxy for the aggregate quantity of capital used. Our first
choice was a depreciation measure, but none was available in WSAAfacts. Hence we
considered using the capital stock variable: “written down current cost of fixed assets”. This
could be a reasonable measure of capital consumption if each firm had a portfolio of assets
with similar average asset lives and hence the stock of capital would be roughly proportional
to the consumption of capital. However the measure we finally decided to use was capital
(CAP) equal to “total expenditure” (TOTEX) minus OPEX. This was because TOTEX was
calculated as OPEX plus capital costs, where capital costs were equal to depreciation plus 4%
of the written down current cost of fixed assets. This measure is clearly designed to be more
a capital cost measure as opposed to a capital quantity measure, however it has the
advantages that it will be affected by average asset lives and it is also the measure which
WSAA members are familiar with.
This capital measure is not without a number of problems. First, it is based on a
depreciated (written down) capital stock measure and hence if a firm has an average asset age
lower than the average firm, they will appear to be using more capital, even though the
service potential of “a kilometre of pipe” is likely to be quite similar across the firms.
Second, different companies use different valuation methods, which is likely to introduce
noise into this measure. Third, the firms tend to do one-off asset revaluations in certain years
(eg. every 5 years or so) and then use the consumer price index (CPI) to revalue their assets
in the intervening years. Given that changes in asset construction costs often deviate from the
CPI (see discussion below), this can mean that a firm which has done a recent revaluation of
assets may appear to be using substantially more (or less) capital relative to the average firm,
depending on the relationship between these two price indices.
The above discussion of the capital measure does not make for happy reading. Hence,
as a robustness check, we have also used the “total length of mains” (MAINS) as an
alternative capital measure in some models. This measure will also be sub-optimal because it
implicitly assumes that the quantities of other capital items (pumps, plants etc.) are used in
proportion to pipeline capital.
When we use data from 1995/96 through to 2002/03 to calculate productivity growth
over time we must also consider how we are going to convert our monetary measures (OPEX
and CAP) into real measures, because they are meant to be proxies for the quantities of inputs
used. In WSAAfacts this issue is dealt with by the use of the CPI. However, the CPI
(reflecting price movements in food, housing, etc.) may be a poor measure of price
movements in water supply variable inputs (eg. labour, chemicals, electricity) and water
supply assets (eg. pipes, pumps, construction services, etc.). To investigate this issue we
searched for some more appropriate price index deflators. Unfortunately, we could find none
that were specific to the water industry, nor to network industries in general. The best indices
we could identify were:
• ABS Catalogue 5204.0, Australian System of National Accounts, Table 8,
EXPENDITURE ON GDP, Implicit Price Deflators, Final Consumption Expenditure,
General Government, State and Local, and
• ABS Catalogue 5204.0, Australian System of National Accounts, Table 8,
EXPENDITURE ON GDP, Implicit Price Deflators, Public Gross Fixed Capital
Formation, Public Corporations, State and Local,
for OPEX and CAP, respectively.18
These two indices, which we will call the OPEX deflator (OD) and the CAP deflator
(CD) are plotted in Figure 1, along with the CPI. Note that the new OPEX price deflator
follows a similar pattern to the CPI, but at a higher level, increasing by 25% as opposed to
18%. The new CAPEX price deflator, however, is well below these two indices, and in fact
falls slightly, by 6%. The flat nature of the new CAPEX price deflator is most likely a
reflection of productivity savings in capital construction over this period.
[Fig 1 here]
To illustrate the effect of the use of these alternative deflators upon the OPEX and
capital measures, we have plotted indices of the various measures (aggregate for the industry)
in Figure 2. The variables involving the CPI deflator are called OPEX and CAP, while those
involving the new deflators are called OPEXN and CAPN. It is interesting to note that when
the CPI is used, the CAP index has no net change over the eight year period, while when the
new capital deflator is used CAPN increases by 25%. Given that the number of connections
has increased by 14% and the water quality requirements have increased over this period, the
CAPN measure is likely to be closer to the “truth” relative to the CAP measure. However,
18 For details, see the ABS web site http://www.abs.gov.au/.
when we observe that the length of mains have only increase by 5%, we begin to suspect that
some number in the middle of 0% and 25% is likely to be closer to the mark.19
[Fig 2 here]
Given the above discussion, we have decided to use length of mains as a capital proxy in our
preferred DEA model. Thus it contains two output variables, WDEL and CONN, and two
input variables, OPEX and MAINS. Technical efficiency (TE), scale efficiency (SE) and
CRS technical efficiency (CRSTE=TE×SE) scores are listed in Table 2 and plotted in Figure
3. The mean TE score is 0.904, which indicates that the average firm could reduce input
usage by 9.6% and still produce the same output level.20 Seven firms have TE scores of 1,
indicating that they are on the DEA frontier: Brisbane, City West, Gosford, Goulburn Valley,
Melbourne, Darwin and Sydney.
The location of a firm such as Darwin on the DEA frontier may come as a surprise to
some, given that it traditionally has high costs per connection (see for example WSAA,
2001). But it should be kept in mind that Darwin has a high WDEL/CONN ratio relative to
most firms, and hence is likely to be located near the fringe of the DEA frontier, with few
other peers to compare with. Similar comments could be made with regard to other firms in
the sample that are especially unique in some aspects. For example, if the are especially large
firms, such as Sydney and Melbourne, relative to the remainder of the sample. Thus, the
DEA method can be a bit too generous to these types of “fringe firms”.21
One way in which the study could be amended to attempt to address this problem
associated with “fringe firms” is to include data on extra businesses from other countries, as
is done in the Carrington et al (2001) gas supply study. The inclusion of data on firms from
other countries can also be important for those firms in the “centre” of the data set if the local
firms are as a group inefficient relative to world’s best practice.22 However, this can be a
19 This discussion emphasises the questionable nature of all the available capital measures, and indicates that
substantially better data would need to be collected before this type of analysis could provide input to a
20 Keeping in mind all previous comments about data quality and model simplifications.
21 Parametric methods, such as SFA, are less susceptible to this type of problem, because the parametric frontier
does not have the degree of local flexibility that a DEA frontier has.
22 For example, see the international benchmarking study of Australian electricity supply in BIE (1996, p96),
where it was found that the performance of the Australian electricity supply industry (measured using a total
factor productivity index) was approximately 30% below the US electricity supply industry in the early 1990’s.
challenging exercise, with data comparability issues generally becoming more complex, as
discussed in Coelli et al (2003, p94).
The mean SE score in Table 2 is 0.903, indicating that the average firm should be able
to reduce input use (per unit of output) by 9.7% if it was able to change its scale of operation.
However, it should be kept in mind that the size of many of these firms is determined by
geographical factors, and hence the option of changing scale is not available. The final
column in Table 2 provides information on the nature of scale economies, from which we
note that all 12 firms that have scale inefficiency do so because of their small size. That is,
they are located on the increasing returns to scale (IRS) portion of the DEA frontier. The
small firms from regional Victoria, Barwon, Central Gippsland, Central Highlands, Coliban
and Goulburn Valley, have the lowest SE scores in the sample.
When we look at the CRSTE scores, which equal the product of the TE and SE
scores, we observe that the CRSTE scores are quite low for these small regional firms. This
is also evident in many of the commonly reported partial productivity measures, such as
OPEX/WDEL and OPEX/CONN, and emphasises the point that these partial ratios can be
quite misleading because they do not account for scale economies.
[Table 2 here]
[Figure 3 here]
Given the concerns that we have with our capital measure, we decided that it would
be wise to repeat the DEA analysis with our MAINS measure replaced with CAP. The
results obtained were reassuringly similar, with the exception of some small changes for
Brisbane and Hunter. The TE scores for the two models are plotted in Figure 4 for ease of
[Figure 4 here]
Another test of the worth of our DEA model is to run a second-stage regression of the
TE scores upon various factors that we know are not explicitly accounted for in the model
and hence may be influencing the TE scores obtained. Hence, using the TE scores from
Table 2 and information on percentage of non-residential connections; percentage of water
from non-catchment sources; average annual rainfall (mm); average maximum temperature
(degrees C); peak to average flow; and electricity consumption per connection (all from
Table 1) we ran a second-stage regression. None of these factors had estimated coefficients
that were significant at the five percent or ten percent levels. Hence, we can be reasonably
confident in the quality of our DEA model.
In the above efficiency analysis we consider two different models because of our concerns
with the capital measure. In our analysis of productivity growth we have the additional
complexity of the choice of price deflators. As a result, we have chosen to calculate four
different sets of Malmquist DEA TFP measures. The technical efficiency change (TEC),
technical change (TC) and TFP change (TFPC) from these four models are summarised in
Table 3. The first set of results relate to the preferred model where MAINS are used to proxy
capital and the new deflator is used to deflate OPEX. For this model we see that average
annual TFP change over these 18 firms over 8 years is equal to a 1.2% decline per year. This
measure can be decomposed into a 2.2% technical regress per year and 1.2% increase in
technical efficiency per year.23
[Table 3 here]
The other three sets of results in Table 3 are within 0.6% of the above TFP change
measures. The second model, involving MAINS and the CPI deflator, obtains TFP change of
minus 1.5% per year. This is slightly below the preferred model results, because the smaller
CPI deflator suggests that OPEX is growing at a faster real rate. The third model, involving
CAPN and the new deflators, obtains TFP change of minus 1.7% per year. This again is
slightly below the preferred model results, because the CAPN measure grows at a faster rate
relative to MAINS, suggesting greater capital input usage. Finally, The fourth model,
involving CAP and the CPI deflator, obtains TFP change of minus 0.6% per year. This is
slightly above the preferred model results, because the CAP measure grows at a slower rate
relative to MAINS, suggesting less capital input usage.
More detailed results for the preferred model involving MAINS and the new deflator are
provided in Table 4, where the time-wise patterns are listed, and in Table 5, where the firm-
23 These figures do not add to zero because of rounding.
level results are provided. The average annual TFP changes vary from plus 3.6% to minus
5.1%, while the average firm-level changes vary from plus 1.6% to minus 5.0%.
[Table 4 here]
[Table 5 here]
The negative measures of TFP change obtained warrant further comment. First, the
price deflators used are approximate, and hence these could be influencing things. Second,
during this period, demand management strategies were put in place in many firms, which
would have had a dampening effect upon WDEL and hence upon the aggregate output
measure. Third, quality improvement strategies were put in place in many firms during this
period, which would be reflected in higher input use, but not in higher output production,
because the quality of the services provided are not explicitly reflected in the DEA model
used. Fourth, we note that some of the smallest companies in the sample have the lowest
productivity growth. That is, Barwon, Central Highlands, Coliban, Goulburn Valley and
Darwin have the lowest TFP growth. Thus our unweighted measure of TFP growth is likely
to understate the aggregate TFP growth in the industry. Lastly, the largest reduction in TFP
growth occurred in the final year of the sample. If we had finished our sample one year
earlier, the average TFP growth would have been almost one percentage point higher.
With reference to some of the above comments, we conducted a few extra
calculations to gauge the sensitivity of our results to some of these factors. First, we
calculated the weighted average TFP growth for the industry using CONN as the weight, and
found that the aggregate TFP change measure increased from minus 1.2% to 0.0%, reflecting
the better performance of the larger firms in the sample. Second, we reran the preferred
model with WDEL omitted from the output set, to attempt to adjust for the possible effects of
demand management initiatives, and obtained an average TFP growth of plus 0.4% per year.
Furthermore, when we applied the above firm weights to these new scores we obtained an
average TFP growth of plus 1.1% per year. However, this measure could potentially
overstate the rate of TFP growth because it does not reflect the fact that WDEL/CONN is
reducing over time, which should imply less cost per connection.
Use in price regulation
How might a regulator use these efficiency and productivity growth results in implementing
price-cap regulation? Given that the regulator is reasonably confident in the data that is used
in the study (which would not be the case in this particular case), we provide the following
In most cases, price caps are set over a five-year term. The regulator will allow firms
to increase prices each year by CPI-X, where X is a measure of the expected productivity
improvements. The value of X is usually based upon a measure of previous TFP growth in
the industry. Also, if the regulator believes some firms are more inefficient than other firms
it will ask the more inefficient firms to achieve higher X factors.
The regulator may require all firms to achieve the weighted mean annual productivity
growth of 1.1 % we obtained from the TFP model where WDEL was omitted (assuming that
demand management policies are likely to continue over the next five years). Furthermore, it
could require firms with DEA technical efficiency scores below one to catch up 50% of the
way to the frontier over the next five years. This is a conservative request, designed to
account for the fact that the technical efficiency scores are measured with error, and also to
reflect the fact that some firms will find it difficult to make efficiency savings if they face
constraints, such as having excess capacity in areas where projected growth is low, etc.
We have used the above rules to construct illustrative X factors for the 18 WSAA
firms. We have taken the TE scores from Table 2 and produced X factors for each firm,
which are reported in Table 6. To illustrate how the X factor values in Table 6 were
calculated, consider the first listed firm, Canberra, which has a TE score of 0.755. It would
be required to catch up (1-0.755)/2=0.123 or 12.3% over the five year period. Which is
(1.16)1/5=1.023 or 2.3% compounded catch-up per year. Thus its X factor would be
1.1+2.3=3.4% per year.
The X factors in Table 6 range from 1.1% per year for the frontier firms, to 4.6 % per
year for Central Highlands, the firm with the lowest technical efficiency score (0.627). The
average X factor is 2.0 % per year. An X factor of 2.0 % implies that the firm must reduce
unit costs in real terms by 2.0 % per year.
However, it should be emphasised that these types of X factors, derived from a process
such as this, should not be used in a prescriptive or mechanical manner. The numbers should
ideally be used as a “starting point” for discussions between the regulator and the regulated.
For example, Adelaide, which has an illustrative X factor of 2.6, that is above the average
value of 2.0, may wish to argue that the DEA model has not properly taken account of the
fact that it must pipe almost half of its water from the Murray river, which results in higher
pumping costs and treatment costs (due to silt and salt) relative to a firm such as Sydney
which derives almost all its water from catchment sources. Adelaide may wish to attempt to
cost out the extra expenses involved and then make a case to the regulator for a reduced X
factor on this basis.
[Table 6 here]
We should also reiterate our earlier comments regarding data quality. The above
illustrative X-factors are based upon a DEA model that used length of mains as a proxy for
the capital input. This is a sub-optimal measure, which was used because we had even less
faith in the reliability of the capital measures, which were based upon a variety of valuation
techniques in different businesses, including (i) a detailed replacement cost valuation of each
item in the asset register, (ii) using the CPI to scale an asset valuation made some years
before, and (iii) in some cases simply scaling the historical cost valuation by a “ball-park”
factor, such as 1.5.
The capital valuation issue is not our only concern. In addition we note that all of these
businesses also supply wastewater services to their customers. To our knowledge, it is
unlikely that all businesses are using the same set of overhead cost allocation rules. Thus it is
possible that some firms may putting more (or less) overhead costs into the “water supply
costs bucket” versus the “wastewater services bucket”, relative to the industry average. If
this is the case, the water supply efficiency measures will be biased.
Furthermore, it should be noted that our initial plan in this study was to attempt to
measure the efficiency of the water distribution business alone. That is, with the wholesale
part of the business (water collection and treatment) removed so we could have a better
chance of comparing like with like, because the wholesale activities are the ones that are most
heavily affected by differing local environmental conditions. However, the published data
did not allow us to do this. This is one avenue that regulators could consider in the future,
when collecting data for exercises such as these.
While on the topic of differing environments, it is worth noting that some observers have
expressed concerns regarding the fact that the above (illustrative) X-factors are based upon a
TE score from a particular year, and that these TE scores can be significantly affected by
annual climatic differences in a country such as Australia, where a large percentage of water
is used on gardens. One possible solution to this problem is to use an average of TE scores
over recent years. However, this could disadvantage those firms who have had TE scores
that are trending upwards in recent years. A preferable option could be to use some form of
smoothing on the volume data, such as using a three-year moving average in the DEA model,
and then simply use the TE score from the final period.
In this study we provide (to our knowledge) the first published set of comprehensive
performance measures for the Australian water supply industry. We use DEA to provide
measures of technical efficiency and scale efficiency for each of the 18 WSAA businesses in
2002/03. We also provide TFP change measures for each firm over the eight-year period
from 1995/96 to 2002/03. Our results indicate that the average firm has a technical efficiency
score of 90.4% and has annual average TFP growth of between minus 1.7 % and plus 1.1%,
depending upon the measures used in each DEA TFP model.
The above range of TFP measures illustrate the importance of the choice of data used
in these studies. Our analysis has highlighted a number of data related issues that warrant
further attention before the results of a study such as this could be considered for use in
aiding the decision making process in the price regulation of water supply businesses. In
particular, the available data on capital needs improvement. WSAA firms are using capital
valuation methods that satisfy the relevant accounting standards. However, the variety of
methods used means that the available data is not comparable across firms. Furthermore, a
lack of appropriate water industry price deflators for use in the TFP calculations is an
additional concern. The ABS deflators used in this study were an improvement over the use
of the CPI, but much work remains to be done in this area.
It should be emphasised that these data problems would apply equally if we were to
use a less sophisticated performance measurement technique, such as OPEX per ML of water
delivered. However, the DEA methods used here have advantages over this type of partial
productivity ratio in that they are able to make adjustments for scale of operations, average
customer size and density, so as to allow more appropriate comparisons of performance.
This study represents our first attempt at the calculation of comprehensive
performance measures for this industry. Various avenues for further work remain. First and
foremost, the analysis should be repeated once better data is obtained (on capital value, price
deflators, etc.). Second, the work could be repeated using stochastic frontier methods (SFA)
to judge the sensitivity of results to the choice of methodology. Third, this study has
focussed on water supply activities. One could repeat the exercise for the wastewater
activities of these businesses, to obtain an indication of the overall performance of the urban
water services industry in Australia.
Anwandter, L., and O. Teofilo, Jr. (2002), “Can Public Sector Reforms Improve the
Efficiency of Public Water Utilities?” Environment and Development Economics, 7,
Ashton, J.K. (2000), “Cost Efficiency in the UK Water and Sewerage Industry.” Applied
Economics Letters, 7, 455-58.
Bhattacharyya, A., T.R. Harris, R. Narayanan, and K. Raffie (1995), “Specification and
Estimation of the Effect of Ownership on the Economic Efficiency of the Water
Utilities”, Regional Science and Urban Economics, 25, 759-84.
Bhattacharyya, A., T.R. Harris, R. Narayanan, and K. Raffie (1995), “Technical Efficiency of
Rural Water Utilities” Journal of Agricultural and Resource Economics, 20, 373-91.
Bhattarcharyya, A., E. Parker and K. Raffie (1994), “An Examination of the Effect of
Ownership on the Relative Efficiency of Public and Private Water Utilities”, Land
Economics, 70, 197-209.
BIE (1996), Electricity 1996: International Benchmarking, Bureau of Industry Economics,
Report 96/16, Canberra.
Bruggink, T.H., (1982), “Public versus Regulated Private Enterprise in the Municipal Water
Industry: A Comparison of Operating Costs”, Quarterly Review of Economics and
Business, 22, 111-125.
Byrnes, P., S. Grosskopf and K. Hayes (1986), “Efficiency and Ownership: Further
Evidence,” Review of Economics and Statistics, 68, 337-341.
Carrington, R., T.J. Coelli and E. Groom (2002), “International Benchmarking for Monopoly
Price Regulation: The Case of Australian Gas Distribution”, Journal of Regulatory
Economics, 21, 191-216.
Coelli, T.J. (1995), “Recent Developments in Frontier Modelling and Efficiency
Measurement”, Australian Journal of Agricultural Economics, 39, 219-246.
Coelli, T.J., A. Estache, S. Perelman and L. Trujillo (2003), A Primer on Efficiency
Measurement for Utilities and Transport Regulators, World Bank Institute,
Coelli, T.J., D.S.P. Rao, and G.E. Battese (1998), An Introduction to Efficiency and
Productivity Analysis, Kluwer Academic Publishers, Boston.
Crain, W.M. and A. Zardkoohi (1978), “A Test of the Property Rights Theory of the Firm:
Water Utilities in the United States”, Journal of Law and Economics, 21, 395-408.
Cubbin, J. and G. Tzanidakis (1998), “Regression versus Data Envelopment Analysis for
Efficiency Measurement: An Application to the England and Wales Regulated Water
Industry”, Utilities Policy, 7, 75–85.
Diewert, E. and D. Lawrence (2004), “Regulating Electricity Networks: The ABC of Setting
X in New Zealand”, invited paper presented at the Asia Pacific Productivity
Conference, July 14-16, University of Queensland, Australia.
Estache, A.R. and A. Martin (2002), “How Different Is the Efficiency of Public and Private
Water Companies in Asia?” World Bank Economic Review, 16, 139-48.
Färe, R., S. Grosskopf and P. Roos (1998), “Malmquist Productivity Indexes: A Survey of
Theory and Practice”, In R. Färe, S. Grosskopf and R.R. Russell (Eds.), Index
Numbers: Essays in Honour of Sten Malmquist, Kluwer Academic Publishers,
Färe, R., S. Grosskopf, M. Norris and Z. Zhang (1994), “Productivity Growth, Technical
Progress, and Efficiency Changes in Industrialised Countries”, American Economic
Review, 84, 66-83.
Feigenbaum, S. and R. Teeples, (1983), Public versus Private Water Delivery: A Hedonic
Cost Approach, Review of Economics and Statistics, 65, 672-678.
Garcia, S. and A. Thomas (2001), “The Structure of Municipal Water Supply Costs :
Application to a Panel of French Local Communities”, Journal of Productivity
Analysis, 16, 5-29.
King, S. and R. Maddock (1996), Unlocking the Infrastructure: The Reform of Public
Utilities in Australia, Allen & Unwin, Sydney.
Lambert, D. K. and D. Dichev (1993), “Ownership and Sources of Inefficiency in the
Provision of Water Services”, Water Resources Research, 29, 1573-1578.
Lynk, E.L. (1993), “Privatisation, Joint Production and the Comparative Efficiencies of
Private and Public Ownership: The UK Water Industry Case”, Fiscal Studies, 14, 98-
Saal, D. and D. Parker (2000) “The Impact of Privatisation and Regulation on the Water and
Sewerage Industry in England and Wales: A Translog Cost Function Approach”
Managerial and Decision Economics, 21, 253-268.
Saal, D. and D. Parker (2001) “Productivity and Price Performance in the Privatised Water
and Sewerage Companies of England and Wales”, Journal of Regulatory Economics,
Teeples, R. and D. Glyer, (1987), “Cost of Water Delivery Systems: Specification and
Ownership Effects”, Review of Economics and Statistics, 69, 399-407.
Thanassoulis, E. (2000a), “The use of Data Envelopment Analysis in the Regulation of UK
Water Utilites: Water Distribution”, European Journal of Operations Research, 126,
Thanassoulis, E. (2000b), “DEA and its Use in the Regulation of Water Companies”,
European Journal of Operations Research, 127, 1-13.
WSAA (2001), WSAAfacts 2001, Water Services Association of Australia, Melbourne,
WSAA (2003), WSAAfacts 2003, Water Services Association of Australia, Melbourne,
Table 1: Descriptive Statistics
Business number of
per km of
Canberra 133 422 44.87 6.77 0.00 584 21 2.27 18
Barwon 118 351 35.79 9.32 99.47 496 20 2.01 19
Brisbane 403 411 68.39 8.19 0.00 1014 25 2.34 264
Central Gippsland 57 360 29.95 10.53 20.08 744 21 3.00 236
Central Highlands 53 336 25.99 9.43 3.68 616 19 1.98 31
City West M 286 424 75.56 10.14 0.00 598 21 1.81 *
Coliban 60 436 31.07 10.00 0.35 476 22 2.21 *
Gold Coast 197 293 69.49 5.08 0.00 1215 26 1.45 85
Gosford 65 256 69.44 4.62 32.83 1268 24 2.00 197
Goulburn Valley 49 621 29.66 12.24 5.10 458 22 2.04 215
Hunter 205 375 46.44 7.80 57.04 1114 22 1.79 169
Melbourne Cons 1472 325 67.49 8.36 0.00 598 21 1.81 *
Darwin 42 838 34.20 11.90 6.90 1953 33 1.52 292
Adelaide 481 372 55.32 5.61 43.67 555 23 2.24 332
South East M. 572 297 70.37 8.22 0.00 598 21 1.93 12
Sydney 1638 388 79.93 7.08 1.36 1186 23 1.51 99
Perth 621 342 52.50 10.95 50.19 781 25 1.73 167
Yarra Valley M 614 306 71.03 7.65 0.00 598 21 1.88 6
Average 393 397 53.19 8.55 17.82 825 23 1.97 143
1. A “*” indicates missing values.
2. All data from 2002/03 except for rainfall and temperature which are 10 year averages and non-catchment, peak and electricity which are from 2000/01.
3. Business names have been altered to more clearly reflect the cities they serve.
Figure 1: Alternative Price Indices, 1995/96 to 2002/03
Figure 2: Industry-level Inputs and Outputs, 1995/96 to 2002/03
95/96 96/97 97/98 98/99 99/00 00/01 01/02 02/03
95/96 96/97 97/98 98/99 99/00 00/01 01/02 02/03
Table 2: DEA efficiency scores, 2002/03
firm TE-CRS TE SE* scale**
Canberra 0.708 0.755 0.937 irs
Barwon 0.618 0.826 0.748 irs
Brisbane 1.000 1.000 1.000 –
Central Gippsland 0.459 0.760 0.604 irs
Central Highlands 0.371 0.627 0.591 irs
City West M 1.000 1.000 1.000 –
Coliban 0.600 0.849 0.707 irs
Gold Coast 0.978 0.999 0.979 irs
Gosford 0.934 1.000 0.934 irs
Goulburn Valley 0.841 1.000 0.841 irs
Hunter 0.749 0.797 0.939 irs
Melbourne Cons 1.000 1.000 1.000 –
Darwin 1.000 1.000 1.000 –
Adelaide 0.847 0.847 1.000 –
South East M. 0.971 0.976 0.995 irs
Sydney 1.000 1.000 1.000 –
Perth 0.832 0.846 0.984 irs
Yarra Valley M 0.984 0.989 0.995 irs
mean 0.827 0.904 0.903
Figure 3: DEA efficiency scores, 2002/03
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
CRSTE TE SE
Figure 4: Comparison of DEA efficiency scores from models using CAP versus
Table 3: Summary of results for four TFP models, 1995/96 to 2002/03
Model TEC TC TFPC
MAINS & new deflators 1.011 0.978 0.988
MAINS & CPI 1.010 0.975 0.985
CAP & new deflators 0.992 0.991 0.983
CAP & CPI 0.991 1.002 0.994
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Table 4: Annual average TFP results, 1995/96 to 2002/03
year TEC TC TFPC
1996/97 0.997 1.039 1.036
1997/98 1.123 0.88 0.988
1998/99 0.993 0.979 0.972
1999/00 0.955 1.01 0.964
2000/01 1.024 1.005 1.029
2001/02 0.991 0.993 0.984
2002/03 1 0.949 0.949
mean 1.011 0.978 0.988
Table 5: Annual average firm-level TFP results, 1995/96 to 2002/03
firm TEC TC TFPC
Canberra 1.040 0.960 0.998
Barwon 1.005 0.963 0.968
Brisbane 1.021 0.974 0.995
Central Gippsland 1.011 0.989 1.000
Central Highlands 0.960 1.016 0.975
City West M 1.000 0.977 0.977
Coliban 1.007 0.943 0.950
Gold Coast 1.008 0.982 0.990
Gosford 1.021 0.986 1.007
Goulburn Valley 0.992 0.962 0.954
Hunter 1.024 0.963 0.986
Melbourne Cons 1.007 0.991 0.998
Darwin 1.000 0.971 0.971
Adelaide 1.039 0.969 1.007
South East M. 1.012 0.996 1.009
Sydney 1.018 0.998 1.016
Perth 1.029 0.968 0.996
Yarra Valley M 0.998 1.001 0.998
mean 1.011 0.978 0.988
Table 6: Illustrative calculation of X factors
firm TE TFPC catch up X factor
Canberra 0.755 1.1 2.3 3.4
Barwon 0.826 1.1 1.7 2.8
Brisbane 1.000 1.1 0.0 1.1
Central Gippsland 0.760 1.1 2.3 3.4
Central Highlands 0.627 1.1 3.5 4.6
City West M 1.000 1.1 0.0 1.1
Coliban 0.849 1.1 1.5 2.6
Gold Coast 0.999 1.1 0.0 1.1
Gosford 1.000 1.1 0.0 1.1
Goulburn Valley 1.000 1.1 0.0 1.1
Hunter 0.797 1.1 2.0 3.1
Melbourne Cons 1.000 1.1 0.0 1.1
Darwin 1.000 1.1 0.0 1.1
Adelaide 0.847 1.1 1.5 2.6
South East M. 0.976 1.1 0.2 1.3
Sydney 1.000 1.1 0.0 1.1
Perth 0.846 1.1 1.5 2.6
Yarra Valley M 0.989 1.1 0.1 1.2
mean 0.904 1.1 0.9 2.0