Victorian Urban Water Utility Benchmarking

Transcription

1 Victorian Urban Water Utility Benchmarking Report prepared for the Essential Services Commission 31 July 2014 Michael Cunningham Economic Insights Pty Ltd 10 By St, Eden, NSW 2551, AUSTRALIA Ph or denis@economicinsights.com.au WEB ABN

2 CONTENTS EXECUTIVE SUMMARY... iii 1 INTRODUCTION Terms of reference Outline of the report Economic Insights experience and consultants qualifications OVERVIEW OF PRODUCTIVITY BENCHMARKING Relevant Literature Productivity Measurement & Regulation DATA & VARIABLES Data Sources Utilities in the Sample Outputs, Inputs & Business Conditions Variables Outputs Inputs Business conditions Summary METHODOLOGIES Multilateral TFP indexes Frontier Analysis Distance Functions Malmquist TFP Index Stochastic Frontier & Random Effects Models RESULTS Index Analysis Stochastic Frontier Analysis Econometric results Productivity trends by utility type Decomposition of productivity trends Comparative trends for Victorian water businesses i

3 5.4 Comparative Technical Efficiency Limitations Discussion of Findings CONCLUSION APPENDIX A: DISTANCE FUNCTIONS, TFP & DECOMPOSITION APPENDIX B: RANDOM EFFECTS ECONOMETRIC RESULTS APPENDIX C: VICTORIAN WATER UTILITIES PRODUCTIVITY REFERENCES ii

4 EXECUTIVE SUMMARY The Essential Services Commission (ESC) engaged Economic Insights Pty Ltd to undertake a productivity analysis of Victoria s urban water utilities, and to benchmark these businesses against urban water utilities in other states and territories using a methodology similar to ESC (2012) An Analysis of the Productivity of the Victorian Water Industry. This study updates that earlier study with three years of additional data. This substantially increases the sample size from 409 annual observations in the previous study covering 54 water businesses, to 596 observations covering 62 water businesses. Notwithstanding the substantial change in the size of the data sample, the results of this study are largely consistent with those of the 2012 study. The main method of analysis used in the study is econometric stochastic frontier analysis, using a translog input-oriented distance function. This is supported by Multilateral TFP analysis. Separate models are estimated using two different measures of capital inputs and the results are averaged. Over the period 2006 to 2013, the productivity of Victoria s four major water distribution businesses decreased at an average annual rate of 0.4 per cent. However, the productivity of Victoria s regional water businesses decreased at an average annual rate of 1.1 per cent over the same period. Interstate water businesses generally had declining productivity over the same period. The productivity of the major interstate utilities is estimated to have decreased at an average annual rate of 1.0 per cent, and the productivity of interstate regional water businesses decreased at an average annual rate of 1.3 per cent over the same period. Decomposition of productivity changes into the effects of technical change, returns-to-scale and changes in business conditions shows that: the main source of declining productivity rates was an increasing degree of technical inefficiency in the industry the effects of technical change were slightly negative in the period 2006 to 2013, but slightly positive over the whole sample period from 1998 to increasing returns-to-scale contributed a small improvement in productivity due to output growth, while changes in business conditions had negligible effect on productivity. The results for individual Victorian water businesses indicate that the three metropolitan water distribution businesses had either minor gains or minor declines in productivity over the period 2006 to 2013, and overall their productivity was relatively static. This contrasts with the regional water businesses of which only two out of 13 had positive productivity growth over the same period. Several regional water businesses had rates of productivity decline exceeding 1 per cent per year. The analysis of comparative productivity levels of Victorian water businesses against their peers throughout Australia indicates: iii

5 The Victorian water utilities estimated to be in the highest quartile of technical efficiency include Westernport Water, South East Water and Lower Murray Water. The Victorian water businesses estimated to be in the middle quartiles include: Yarra Valley Water, South Gippsland Water, GWMWater, City West Water, East Gippsland Water, Western Water, and Goulburn Valley Water. Those estimated to be in the lowest quartile of technical efficiency include Gippsland Water, Wannon Water, Barwon Water, Central Highlands Water, Coliban Water and North East Water. iv

6 1 INTRODUCTION This report presents the findings of a benchmarking study of the productivity performance of Victorian water businesses in comparison to utilities throughout Australia. Benchmarking refers to methods of identifying the efficient cost of supply or the feasible level of technical efficiency by comparing the performance of several businesses and ascertaining the highest levels of efficiency achieved. This study focuses on benchmarking total factor productivity (TFP) performance. The TFP performance of network industries is of considerable interest to both managers and regulators. As a comprehensive measure of overall economic performance, TFP can provide managers with important information on the overall performance of their business from one year to the next. It enables targets to be set for productivity growth and its progress to be monitored. This provides managers and owners of GDBs with a ready means of gauging the success of reform efforts. Measurement of industry level and firm-specific TFP performance is of interest to regulators seeking to determine price outcomes that are consistent with competitive market outcomes in an industry operating under natural monopoly conditions. Information from industry and firm-level TFP studies can be used when setting X factors in CPI X regulation. It also provides the regulator with a means of assessing whether available efficiency improvements have been achieved during the past regulatory period and may provide insights into what further efficiency improvements are available in the forecast period. 1.1 Terms of reference The Essential Services Commission (ESC) has engaged Economic Insights Pty Ltd ( Economic Insights ) to undertake a productivity analysis of Victoria s urban water utilities, and to benchmark these businesses against urban water utilities in other states and territories. The methodology is similar to that used in the ESC (2012) report: An Analysis of the Productivity of the Victorian Water Industry. 1 This study updates the data used previously to include information in the National Water Commission s (NWC) recently released National performance report : Urban water utilities. The study calculates index-number productivity measures and undertakes econometric productivity analysis in order to estimate, for each Victorian urban water utility and for selected urban water utilities in other jurisdictions, the productivity trends and comparative productivity levels. The methodologies used in the ESC s previous study, and in this study, include: productivity indexes, and stochastic frontier and random effects econometric analysis. Productivity trends are decomposed into separate effects relating to changes in technical 1 1

7 efficiency, technical change, economies of scale, and changes in business conditions. Possible reasons are suggested for differences in productivity trends and levels between groups of businesses, such as major urban versus regional and Victorian versus interstate. 1.2 Outline of the report This report is structured as follows: Section 2 briefly discusses the purposes of productivity benchmarking in natural monopoly regulation and provides a short introduction to the Australian literature relating to productivity analysis of the water industry. The data set used in this study is described in section 3. It is primarily based on National Performance Report for urban water utilities published by the National Water Commission (NWC), and on earlier statistical reports published by the Water Services Association of Australia (WSAA). Section 3 also explains the choice and definition of the outputs, inputs and other variables. This study uses two distinct methodologies, which are explained in section 4. One of these is the multilateral index number method of estimating the comparative productivity levels of utilities over time. The second methodology is an econometric analysis of inefficiency using distance functions, which is also used to identify the rates of productivity change for each utility as well as comparative measures of technical efficiency. It also permits productivity change to be decomposed into several sources, including: technology improvement; reduction in technical inefficiency; and returns to scale effects. Section 5 presents the results of the multilateral index number analysis and the stochastic frontier analysis. The main findings in relation to productivity trends and comparative productivity levels of different utilities are summarised. Productivity trends are decomposed into the effects of technical change, returns-to-scale effects, changes in business conditions and changes in the degree of technical inefficiency. Section 6 briefly provides a concluding summary on the main findings and observations made in this report. Care needs to be taken when interpreting the results of this analysis in light of limitations in the quality and completeness of the data set. Although the WSAA and NWC data is the best available, it is dependent on the quality of information reported by water utilities, which may be variable. In some instances of incompleteness, interpolation has been necessary. Reviewing and enhancing the quality of summary data would be a valuable exercise for the Victorian water industry. 1.3 Economic Insights experience and consultants qualifications Economic Insights has been operating in Australia for 20 years as an infrastructure consulting firm. Economic Insights provides strategic policy advice and rigorous quantitative research to 2

8 industry and government. Economic Insights experience and expertise covers a wide range of economic and industry analysis topics including: infrastructure regulation; productivity measurement; benchmarking of firm and industry performance; infrastructure pricing issues; and analysis of competitive neutrality issues. This report has been prepared by Michael Cunningham who is an Associate of Economic Insights. 3

9 2 OVERVIEW OF PRODUCTIVITY BENCHMARKING This chapter briefly discusses the purposes of productivity benchmarking in the context of natural monopoly regulation and provides a short introduction to the Australian literature relating to productivity analysis of the water industry. Productivity is a measure of the physical output produced from the use of a given quantity of inputs. All enterprises use a range of inputs including labour, capital, land, fuel, materials and services. If the enterprise is not using its inputs as efficiently as possible then there is scope to lower costs through productivity improvements and, hence, lower the prices charged to consumers. This may come about through the use of better quality inputs including a better trained workforce, adoption of technological advances, removal of restrictive work practices and other forms of waste, and better management through a more efficient organisational and institutional structure. When there is scope to improve productivity, this implies there is technical inefficiency. This is not the only source of economic inefficiency. For example, when a different mix of inputs can produce the same output more cheaply, given the prevailing set of inputs prices, there is allocative inefficiency. Productivity is measured by expressing output as a ratio of inputs used. In this study the term productivity refers to Total Factor Productivity (TFP), which is defined as the ratio of an index of all outputs to an index of all inputs used in producing those outputs. 2 The rate of change in TFP can be estimated as the difference between the rates of change of the index of outputs and the index of inputs. Output can be increased by using more inputs, making better use of the current level of inputs and by exploiting economies of scale. The TFP index measures the impact of all the factors effecting growth in output other than changes in input levels. As noted in Lawrence (1992), by providing a means of comparing efficiency levels, TFP measurement is an ideal tool for promoting yardstick competition in non competitive industries. It provides managers with useful information on how their business is performing overall and on how it is performing relative to its peers. TFP measurement, thus, provides a ready means of benchmarking the business s overall performance relative to other businesses supplying similar outputs. 2.1 Relevant Literature Previous studies of water productivity in Australia were summarised briefly by Cunningham (2013) and international studies of water industry productivity and efficiency were reviewed by Abbott & Cohen (2009). The most significant Australian studies that preceded the ESC s appear to be Woodbury & Dollery (2004), Coelli & Walding (2006), Byrnes et al. (2010) and Worthington (2011). The Woodbury & Dollery and Byrnes et al. studies were focussed on New South Wales local government water supply services. The Coelli & Walding study was 2 Partial factor productivity measures are not examined in this study. These measures express one or more outputs relative to one particular input (eg labour productivity is the ratio of output to labour input). 4

10 undertaken on behalf of the Water Services Association of Australia (WSAA) and the Worthington study was carried out on behalf of the National Water Commission (NWC). All of the studies mentioned used data envelopment analysis (DEA). The ESC study appears to be the first productivity benchmarking study of the Australian water industry to employ stochastic frontier analysis. The ESC s 2012 study (also published as: Cunningham 2013) adopted a similar methodology to the Saal & Parker (2006) and Saal, Parker & Weyman-Jones (2007) studies of the productivity of the UK water industry. This involved the estimation of an input-oriented distance function using stochastic frontier analysis (see: Coelli et al. 2003). The ESC study also used a TFP index-based approach and an alternative econometric method using the random effects specification to estimate the distance function. The index-based method was primarily used as a cross-check against the other methodologies and the random-effects model was viewed as a useful comparative method (ESC 2012a pp.11-12). The stochastic frontier model was preferred on both conceptual and goodness-of-fit grounds (ESC 2012a p.15). The key findings of the ESC study were: Trends in TFP: From 2006 to 2010 the TFP of Australian urban water distribution businesses declined on average by 0.7 per cent per year. The four largest Victorian water distributors performed better than average, with only a minimal TFP decrease. However, Victorian regional water businesses were amongst the worst performers, with an average annual rate of TFP decrease of 0.8 per cent over the same period. Comparative technical efficiency: The four major urban water utilities were among the more efficient in the sample, and taken together were 13 per cent more efficient than the average water business. Water utilities in regional Victoria were among the least efficient in the sample, and on average were 9 per cent less efficient than the average water utility. More recent productivity studies include the Productivity Commission (PC 2012) study of productivity trends in the electricity, gas and water sectors, and the CIE (2012) study carried out on behalf of the Essential Service Commission of South Australia (ESCOSA). The PC s study was comparatively broad, covering all urban water supply and sewerage services providers, rural water authorities, bulk water suppliers, catchment authorities and schemes for supplying water to farms for irrigation. It was focussed on productivity index trends for the sector as a whole, rather than benchmarking between firms. Outputs of the sector were measured by the Australian Bureau of Statistics (ABS) index of real value added, which the PC showed could be approximated by a combination of three variables: urban water sales; irrigation water sales; and sewerage connections. The PC found that over the period to the productivity of the water, sewerage and drainage industry declined at an average annual rate of 4.3 per cent per year. However, the PC examined the effects on productivity of demand management due to drought, and improvements in the quality of water supply and sewerage services, particularly the levels to which sewage was treated. It estimated that the drought effects and the quality improvements explained about 80 per cent of the productivity decline after (PC 2012 p.109). 5

11 The CIE study was focussed on benchmarking SA Water against other water utilities. It followed the same general approach as the ESC study, but used both Cobb-Douglas and translog specifications for the input distance function and adopted a different measure of capital services (CIE 2012 p.59). 3 While the results for both the Cobb-Douglas and translog specifications were presented, the CIE felt that the latter yielded less intuitive results. Sensitivity analysis was undertaken with different measures of capital inputs, but the preferred approach to capital services measurement differed from that used by the ESC in two main respects: Capital services inputs were calculated using the perpetual inventory method using actual capital expenditure and assumed depreciation of two per cent per year, but the starting points for this calculation were assumed. The initial capital base for each utility was set to a common value and businesses that do not have data in 1997/98 are given a starting capital base equal to the average of the other utilities in the year in which they first have data (CIE 2012 p.38). Capital expenditure related to security of supply, particularly desalination plants, was excluded due to concern that there is no readily available data for changes in the security of supply for urban water utilities over time and without an output measure for security of supply this may downwardly bias productivity measures (CIE 2012 p.37). We are not convinced that either of these two assumptions are an improvement on the ESC s approach. Firstly, the assumption of a common initial capital base does not make use of available information on either the quantities of capital infrastructure used by each of the different water businesses or the valuation of their fixed assets. It is true that when the perpetual inventory method is used over very long periods (e.g. 100 years would be relevant in this context with a 2 per cent declining balance depreciation rate) the starting values for the calculation become unimportant. However, over the relatively short period over which the CIE performed the calculation, that is not the case. Secondly, the CIE noted considerable investment in projects to improve security of supply, including desalination plants and water recycling projects, but there was a lack of evidence that security of supply was increasing over time. A different impression is given by information presented by the PC, indicating that urban dam storage capacity per capita (a useful measure of security of supply prior to desal plants coming on stream in recent years) declined continuously after the mid-1980s, and in 2010 was at its lowest level since 1982 (PC 2012 p.84). This may suggest that security of supply was declining in many places over much of the period being studied. 2.2 Productivity Measurement & Regulation Benchmarking is increasingly being used in the regulation of essential infrastructure services 3 The CIE study also used a different formulation of the stochastic frontier model, with time-invariant inefficiencies, rather than the time-varying decay of inefficiency formula used by the ESC. 6

12 in Australia. The AER (2013b) has indicated it will be making greater use of benchmarking for assessing the efficiency of network businesses. The two forms of benchmarking it intends to use as an integral part of future energy infrastructure price reviews are: benchmarking a network business expenditure when disaggregated into cost categories, termed category analysis, and economic benchmarking of the efficiency of a network business regulatory operations as a whole. The analysis in this report is an application of latter type of economic benchmarking. The benchmarking techniques the AER is likely to have particular regard to include multilateral TFP analysis, data envelopment analysis (DEA) and econometric modelling (AER 2013a). Two of these methods are used in this report. As mentioned in the introduction, the primary purpose of this study is to update the ESC (2012a; b) water productivity study using a similar methodology. That study used a methodology consistent with Coelli et al. (2003) and Saal, Parker, and Weyman-Jones (2007). The ESC study used data from (for larger water businesses) or from (for smaller water businesses) up to , and this study extends the database to More detail on the data and methodologies used is provided in the following chapters. 7

13 3 DATA & VARIABLES This chapter addresses two issues of fundamental importance in benchmarking studies, the available data and the definitions of the outputs and inputs of the businesses in the industry being analysed. The sufficiency and quality of data is perhaps the biggest hurdle to be able to conduct a robust, defensible analysis (Coelli et al p.83). Fortunately, the Water Services Association of Australia (WSAA) and the National Water Commission (NWC) have gathered data for a substantial number of Australian water utilities. Section 3.1 outlines the sources of data used for this study and section 3.2 summarises the utilities and periods included in the database used in the analysis. One of the most difficult aspects of productivity analysis is deciding on how many and which outputs to measure and how they are to be measured. Data limitations and computational constraints require compromises on the number and nature of the outputs and inputs included in the analysis, and these choices have an important influence on measured productivity. The definitions of outputs and inputs used in this study largely follow the ESC s 2012 study. They are explained in section Data Sources This study builds on data collected for the ESC s 2012 report which covered the period from to for major urban water utilities and shorter periods (usually from to ) for smaller utilities. Most of the data used in the 2012 study was sourced from the Water Supply Association of Australia s WSAAfacts reports, and from the National Water Commission s (NWC) National Performance Report: urban water utilities. Data for bulk water prices paid by urban water businesses was sourced directly from bulk water suppliers and regulators. Data for temporary water restriction levels was sourced mainly from local governments in all states. For this study the database has been extended to include the three years from to Most of the additional data included in this study was obtained from the National Performance Report : urban water utilities (NWC 2014). Data for bulk water prices paid by urban water utilities was sourced from company annual reports and regulator determinations and reports. Information on temporary water restrictions was obtained from media releases and news articles. 3.2 Utilities in the Sample In the 2012 study the ESC used a sample of 54 urban water utilities from seven states and territories in Australia, with 409 observations in total. Three of these 54 utilities represented the combined operations of two businesses (separate sewerage and urban water providers) to ensure that all of the utilities in the sample were combined water and sewerage providers. The dataset was an unbalanced panel, since the periods for which data was available differed between water businesses. Since the last study was undertaken, the main changes to data availability (apart from the 8

14 addition of three years of data) relate to the Queensland urban water utilities. Firstly, NWC ceased reporting data for Brisbane Water, Gold Coast Water, Ipswich Water and Logan Water after Since then, the NWC reported data for Queensland Urban Utilities (which replaced Brisbane Water, Ipswich Water and others) but did not report data for Allconnex (which replaced Gold Coast Water, Logan Water and others 4 ). The NWC now also reports data for some additional Queensland urban water businesses, which have been included in the sample, even though there is only three years of data for some of them. Table 3.1 provides a list of all the businesses included in the sample and the year for which data is available for those businesses. The updated sample covers 62 businesses with 596 observations in total, representing an average of 9.6 observations per utility. This represents a substantial increase in the size of the data sample. Since it is an unbalanced panel, care needs to be taken when interpreting average productivity movements over time, because they may be calculated using a changing set of utilities. Table 3.1 Summary of data sample # Utility Sample Period Observations Australian Capital Territory 1 ACTEW 1998 to New South Wales 2 Hunter Water Corporation 1998 to Sydney Water Corporation 1999 to Gosford City Council 1998 to Wyong Shire Council 2006 to Albury City Council 2006 to Coffs Harbour City Council 2006 to MidCoast Water 2006 to Port Macquarie Hastings Council 2006 to Shoalhaven City Council 2006 to Tweed Shire Council 2006 to Wagga Wagga / Riverina Water 2006 to Ballina Shire Council 2006 to Bathurst Regional Council 2006 to Bega Valley Shire Council 2006 to Byron Shire Council 2006 to Clarence Valley Council 2006 to Essential Energy 2006 to Allconnex is now no longer the water utility for the Gold Coast. 9

15 Table 3.1 Summary of data sample (cont.) # Utility Sample Period Observations 19 Dubbo City Council 2006 to Eurobodalla Shire Council 2006 to Goulburn Mulwaree Council 2010 to Kempsey Shire Council 2006 to Lismore City Council 2006 to Orange City Council 2006 to Queanbeyan City Council 2006 to Tamworth Regional Council 2006 to Wingecarribee Shire Council 2006 to Northern Territory 28 Power and Water Darwin 1998 to Power and Water Alice Springs 2006 to Queensland 30 Brisbane Water 1998 to Gold Coast City Council 1998 to 2010 & Cairns Water and Waste 2008 to Fitzroy River Water 2009 to Queensland Urban Utilities 2011 to Unitywater 2011 to Water and Waste Services Mackay 2008 to Wide Bay Water 2011 to Ipswich Water 2002 to Logan Water 2003 to 2010 & South Australia 40 SA Water - Adelaide 1998 to Victoria 41 Barwon Water 1998 to City West Water 1998 to South East Water Ltd 1998 to Yarra Valley Water 1998 to Central Gippsland Water 1999 to Central Highlands Water 1998 to Coliban Water 1998 to Goulburn Valley Water 1998 to East Gippsland Water 2006 to GWMWater 2006 to Lower Murray Water 2006 to

16 Table 3.1 Summary of data sample (cont.) # Utility Sample Period Observations 52 North East Water 2006 to Wannon Water 2006 to Western Water 2005 to South Gippsland Water 2006 to Westernport Water 2006 to Western Australia 57 Water Corporation Perth Metropolitan 1998 to Water Corporation - Mandurah 2006 to Water Corporation - Albany 2006 to Aqwest & Water Corporation Bunbury 2007 to Water Corporation - Geraldton 2011 to Kalgoorlie-Boulder (sewerage) & Water Corp to Total (Avg: 9.6) Outputs, Inputs & Business Conditions Variables The technology of a business is the method of transforming inputs into outputs, and as mentioned, productivity refers to the quantities of outputs that can be produced relative to the quantities of inputs used. In mature industries with moderate technical change most of the businesses in the industry have access to the same or similar technologies. However, the ability to translate inputs into outputs can vary between markets and localities due to external factors such as weather and topography, as well as market characteristics. This is particularly relevant to regulated industries that have monopolies in discrete localities that may have different characteristics. The inputs and outputs used in the analysis refer to the general inputs and outputs used and produced by all urban water utilities, but other variables are also included in the analysis to reflect some of the differences in the business conditions faced by different utilities which will affect the ability of best practice businesses to transform inputs into outputs. The definition of outputs and inputs used in this study largely follow those adopted by the ESC in 2012, because they are suitable and consistent with best practices in water industry productivity studies. For a discussion of different output and input measures that have been, or could be, used in water productivity studies see Berg (2010 app.1) Outputs The ESC s approach to defining outputs was to include customer numbers to represent the site-related and customer-related services including the provision and maintenance of connection to the water supply and sewerage systems and management of customer accounts. Measures of the quantities of water supplied to premises, and of sewage treated, were qualityadjusted because higher quality of drinking water or of sewage treatment requires more inputs to produce than lower quality. Some studies have used performance indicators such as 11

17 customer satisfaction or the level of complaints as outputs, but these may better be viewed as outcomes rather than outputs. Urban water utilities are taken to have the following outputs: (a) the number of customers (the greater of the number of water or sewerage customers, but in almost all instances the former) (b) the quality-adjusted quantity of water supplied to residential and industrial customers, not including water losses, and adjusted for constraints associated with temporary water restrictions (TWRs) (c) the quality-adjusted quantity of sewage treated, including trade-waste. Qualityadjustment is in terms of the levels to which sewage has been treated. The methods of adjusting for quality and for TWRs can be summarised as follows. 5 The adjusted quantity of water supplied to customers, is calculated using the following formula: (3.1) Water supplied DWQ / N where DWQ refers to an index of drinking water quality and N is a normalisation factor to adjust for the effects of temporary water restrictions. DWQ is measured as of the product of: the proportion of zones for which health-related microbiological standards were met the proportion of zones for which health-related chemical standards were met. Adjustment for Temporary Water Restrictions (TWRs) is intended to remove the short-term impacts on productivity caused by these regulatory quantitative constraints. In principle, such constraints can cause under-utilisation of capacity in the short-term until they are withdrawn. The normalisation factor, N, is defined as: (3.2) N = 1 (TWR/ TWR MAX ) [(s 0.25) + (1 s) 0.135] where TWR is the restriction level applying to the customers in the region served by a utility in a particular year; TWR MAX is the maximum restriction stage in each state and territory; and s is the proportion of residential use out of total water use. When TWRs apply, N is less than 1 and when no TWRs apply, N equals 1. Usually at the maximum restriction all outdoor use of water is prohibited. The formula is based on the assumption that outdoor water use represents 25 per cent of residential water use on average, and 13.5 per cent of the water use of non-residential customers. The formula implies that the impact of restrictions is a linear function of the restriction level. The proportions of total water supplied that are used by residential and non-residential customers vary between utilities and over time. The quality-adjusted quantity of sewage treated is defined as: (3.3) Sewerage treated WWQ 5 In other respects the measures are as defined in NWC (2013). 12

18 Where WWQ is an index of sewerage treatment quality measured by the following index: 6 (3.4) WWQ = (% primary 1 + % secondary 2 + % tertiary 3) / 3 If all sewage is treated just to a primary level then WWQ equals When all sewage is treated to a tertiary or advanced level then WWQ equals Inputs This study uses two broad inputs: capital and non-capital inputs. This is consistent with the ESC s 2012 study, and similar to other relevant productivity studies such as the Coelli & Walding (2006) study of the Australian water industry and the Saal et al (2007) study of the UK water industry. In the ESC 2012 study there were two different measures of capital inputs, and each was treated as a separate input. This study closely follows the ESC s approach with regard to defining non-capital inputs and with regard to undertaking the analysis using two different measures of capital inputs. However, the two capital measures are included in alternative econometric models rather than being included in the same model. The results from the two models are then combined. Before explaining the measurement of non-capital and capital inputs in detail it will be useful to first discuss the measurement of inputs in more general terms. There are two general approaches to measuring the quantities of outputs or inputs. The first is direct measurement of those quantities, which is used here for all of the outputs as previously described. The second approach involves using monetary measures, such as revenue or cost, and deflating those measures using an appropriate price index, to obtain an index of the quantities of inputs or outputs. This second approach is often used for certain types of inputs for which quantity data may not be available, or may not reflect the diversity of the inputs within a given class. For example, opex costs usually comprise labour costs, both direct and contracted, and a wide range of materials and services spanning operational consumables, office activities and contracted and in house professional services. In the present case, to derive the implicit quantity of opex inputs we need a price index to deflate annual opex cost, which ideally is a weighted average of the key opex components prices. Non-Capital Inputs The approach to measuring non-capital inputs used here relies on data for nominal operating costs and the use of an appropriate deflator. As defined by the NWC, operating costs do not include depreciation (NWC 2013). The data published by the NWC (and earlier by WSAA) separates operating costs between water supply and sewerage services. But as with the ESC s previous study, we have not treated operating costs for water supply and sewerage services as two separate inputs, in part because cost allocation practices may not be consistent across water businesses. 6 This index was developed by the ESC based on treatment cost relativities sourced from: Ong S & Adams B (1987) A Heuristic Method for Modelling of Treatment Cost Functions, International Journal of Environmental Studies, 29: , p

19 There are three important aspects of the method of developing a quantity index of non-capital inputs used here: adjustment for the Victorian Government environmental levy treatment of purchased bulk water which is included in opex reported by the NWC, and the deflator. Since October 2004, Victorian water businesses have been required to pay an environmental contribution to the Victorian Government. The amounts are specified in regulations, usually amounting to more than five per cent of water supply revenue. The ESC 2012 study excluded this contribution from the operating costs of the Victorian utilities, and the same practice has been adopted here. It is assumed that water utilities in other states do not incur a similar government levy. 7 Operating costs have been separated into the cost of bulk water purchases and all other operating costs. Data for the cost to utilities of purchasing bulk water was gathered from their Annual Reports and from determinations of regulatory agencies. The quantities of purchased bulk water are published by the NWC. The cost of bulk water was separated into fixed and volumetric components. Although it would be desirable to treat bulk water purchases and other non-capital inputs as separate inputs in the econometric analysis, it is not feasible within the translog specification of the distance function because not all utilities purchase bulk water and the logarithm of zero is undefined. As with the ESC s previous study these two non-capital inputs were combined into a composite index of non-capital inputs. The formula used is: (3.5)! NCI it = S i ( B it B) + (1 S i ) R it R ( ) where is the index of non-capital inputs for utility i in period t; is the average share! NCI it! S i of bulk water charges in utility i s total nominal operating costs, and: B represents purchases of raw or treated bulk water in megalitres (ML), and!b is the sample average value of B. R represents the residual operating and maintenance expenditure (not including depreciation) appropriately deflated, and!r is the sample average value of R. The reason for dividing by the sample average values of bulk water quantities and real residual opex is to create unit free measures to facilitate aggregation via equation 3.5. The deflator used for calculating R from the residual operating costs is the average of the 7 In some cases levies may be included within bulk water charges rather than directly imposed on the water distribution business. 14

20 Consumer Price Index, All Groups, Australia 8 and the Wage Price Index (total hourly rates of pay excluding bonuses) for the Electricity Gas Water & Waste sector. 9 Capital Inputs The measurement of capital inputs is often the most problematic aspect of TFP measurement. Reflecting this difficulty, the ESC s study included two different measures of capital inputs in the econometric model that was estimated. One of these was based on the written-down value of fixed assets used in water supply and sewerage services. The other was a quasi-physical measure, largely based on the length of water supply and sewerage mains, but with some adjustment to take into account headworks and desalination plants. It is generally desirable to use physical measures of capital employed to measure the services provided by capital inputs. An important limitation of using monetary values of capital is that they should reflect the present value to the business of those assets over their remaining lives, rather than the current levels of services provided by them. Measures of this kind will decline as an asset ages, because it has a shorter remaining life, even when its current productive contribution remains undiminished. The main benefit of using a monetary measure of capital inputs is that it usually encompasses all of the fixed assets employed. On the other hand, comprehensive information on the physical characteristics of a utility s fixed assets may not be available. For example, the NWC publishes data for the length of water supply and sewerage mains, but this is an incomplete measure that does not take into account a range of other assets employed by a water business. In the present context each measure has strengths and weaknesses. The National Water Commission (NWC) publishes data for the written-down replacement cost of each utility s fixed water supply and sewerage assets, and also publishes capital expenditure data. The asset values have been subject to revision in recent years, which may be associated with the adoption of more consistent accounting practices in the industry. Rather than fully relying on the series of written-down asset values published by the NWC, the ESC study constructed a series based on the value of each utility s fixed assets in the final year of the sample period, and backwardly extrapolating by deducting capital expenditure and adding back depreciation at an assumed uniform depreciation rate. This method has the benefit of removing the effects of stepwise asset revaluations that may have occurred in previous years due to factors such as changes in accounting policies. In this study we have tested the same method of constructing a revised series for the writtendown value of fixed assets. However, we have found only minor differences in the econometric results and productivity trends when this series is used compared to when the NWC s original fixed asset written down value series is used. We have reported the results using the series published by the NWC for the benefit of transparency and simplicity. The deflator used to adjust this series into real terms was the Net Capital Stock deflator of 8 Australian Bureau of Statistics, Consumer Price Index, Australia. 9 Australian Bureau of Statistics, Wage Price Index, Australia, Table 5a. 15

21 Electricity, Gas, Water and Waste industry. 10 The quasi-physical measure of capital used previously was a function of the length of water supply and sewerage mains, the proportion of water that is sourced from a utility s own upstream facilities and groundwater, and the capacity of its desalination plants (if any). In this study we prefer to use the length of water supply and sewerage mains as the physical measure of capital, rather than the quasi-physical units used previously. The reason is that the quasiphysical measure should be viewed as experimental and would not in our view be appropriate for use in regulatory applications until it has been subject to greater scrutiny and comment from stakeholders, particularly in the water industry. In summary, two measures of capital inputs are used as alternatives in this study, a monetary measure based on the written down value of fixed assets and a physical measure based in the length of water supply and sewerage mains. These are used in alternative models, but the average result from both models is reported. The approach used previously of including both measures within the same model makes sense from an econometric perspective, but may raise issues of interpretation from an economic theory perspective, for example in relation to the implied substitutability between the two measures of capital. The approach adopted here of averaging the results from two models has been used in previous econometric productivity studies (eg, Economic Insights 2012). Desalination projects As mentioned in section 2.1, the CIE removed the cost of desalination plants from its measure of the capital stock for the purposes of benchmarking productivity. We have not adopted the same approach here. Table 3.2 shows a list of the six major seawater desalination plants completed in Australia. Only those in Western Australia (WA) and South Australia (SA) are owned by water businesses included in our sample. Of those, the two in WA operate continuously and supply a large proportion of Perth s water use. It would be inappropriate to remove their value from the Water Corporation s fixed assets. The SA plant is currently on standby, waiting to operate when needed. This is the only desalination plant that could adversely affect the productivity estimates. In other contexts it may be desirable to remove it, but that would require a careful examination of Adelaide s past and present security of supply situation, which is beyond the scope of this study. 10 Australian Bureau of Statistics, Australian System of National Accounts, Table 58. Capital Stock, by Industry. 16

22 Table 3.2 Desalination plants Facility Details Owner Use Perth Seawater Completed: 2006 Cost: Water Corporation In continuous operation Desalination Plant $387m Capacity: 45 GL/a Southern Seawater Completed: 2013 Cost: Water Corporation In continuous operation Desalination Plant $1,085m Capacity: 100 GL/a Adelaide Completed: 2011 Cost: SA Water Standby mode Desalination Plant $1,824m Capacity: 100 GL/a Gold Coast Completed: 2009 Cost: Seqwater (owned by Standby mode (until Desalination Plant $1,120m Capacity: 45 GL/a WaterSecure pre-july dam levels are below 2011) 60%) Kurnell Desalination Completed: 2010 Cost: Sydney Desalination Standby mode (until Plant $1,803m Capacity: 90 GL/a Plant Pty Ltd (under 50- dam levels are below year lease from NSW 70%) Government) Victorian Completed: 2012 Cost: Aquasure consortium Standby mode (until Desalination Plant $3,500m Capacity: 150 GL/a (under long-term BOOT dam levels are below arrangement). 65% in March) Business conditions In the 2012 ESC study, three variables representing business conditions were included in the econometric productivity analysis: the proportion of water sourced from groundwater the proportion of customers with sewerage connection, and the proportion of wastewater collected that is trade waste. In this study the number of business conditions variables has been increased in an effort to control for factors that may cause different productivity outcomes, or be related to them. Less emphasis has been given to excluding such variables if found to be statistically insignificant. The additional variables are: the proportion of water sourced from recycled water net greenhouse emissions associated with water supply per household. The reason for including the latter variable is as follows. One of the unfortunate limitations of the available data is the lack of information relating to the use of energy inputs, which may reflect different business conditions. 17

23 For example, topographical factors can have an important influence on the inputs required to supply customers in different areas, such as the amount of energy needed for pumping water or wastewater. Kenway et al (2008) estimated that Sydney Water used 1.8 terajoules per GL of energy in pumping water and wastewater compared to 1.1 TJ/GL for for utilities in Melbourne. The other utilities estimated were 1.7 TJ/GL in Perth, 0.4 in Brisbane, 1.0 for the Gold Coast and 4.4 TJ/GL for Adelaide. (Cunningham 2013 p.184) 11 Including net greenhouse gas emissions from water supply operations as a business condition variable is an attempt to remedy this omission on the assumption that energy use and net greenhouse gas emissions may be correlated Summary This study assumes there are three outputs and three inputs. The outputs are: the number of customers supplied; a measure of water supplied which is both quality-adjusted (for drinking water quality) and normalised for the effect of temporary water restrictions; and the quantity of sewage treated which is quality adjusted (for the sewage treatment level). The inputs comprise capital and non-capital inputs. Non-capital inputs are a composite index of bulk water purchased and all other non-capital inputs. There are two different measures of capital inputs. One of the measures of capital inputs is in physical units, namely the length of water supply and sewerage mains. This method has the shortcoming of being incomplete. It does not include upstream water gathering or groundwater facilities, or water and sewerage treatment capacity. Nor does it include desalination plants, among other things. The other measure of capital inputs is an accounting based measure, using the depreciated replacement cost of assets. This measure has the benefit of completeness, but its main shortcoming is that the value of assets is a wealth-based measure which will vary depending on the age of the assets, and need not closely correlate with the actual services provided by those assets in the current period. In this study a separate model is estimated using each of the two capital measures. This approach differs from the previous study, in which each measure was treated as a separate input and included in a single econometric specification. 11 The reference in the quote is to: Kenway, Priestly, Cook, Seo, Inman, Gregory & Hall (2008) Energy Use in the Provision and Consumption of Urban Water in Australia and New Zealand, CSIRO National Research Flagships & Water Supply Services of Australia. 18

24 4 METHODOLOGIES The ESC has requested Economic Insights to carry out both an index-based analysis of TFP and an econometric analysis. These methodologies are discussed in sections 4.1 and 4.2 respectively. 4.1 Multilateral TFP indexes The ESC (2012) used an index number method for comparing productivity between utilities and trends over time as an alternative to econometric analysis. TFP indexes represent the ratio of an index of all outputs to an index of all inputs, where each index is defined relative to a specific base period. An index number is a unit-free measure, and because it is defined relative to the base period it represents the proportionate change in productivity relative to that base period. The index number method used by the ESC had fixed weights for each input and output and the same weights were applied to all utilities. Although this method was sufficient to make comparisons between utilities as well as over time, and to provide a cross-check against the econometric results, it is not the most suitable index method for this purpose. The multilateral translog TFP (MTFP) index measure developed by Caves, Christensen and Diewert (1982) allows comparisons of the absolute levels as well as growth rates of productivity. It satisfies the technical properties of transitivity and characteristicity which are required to accurately compare TFP levels within panel data. 12 It has been used in TFP analysis in other regulated sectors in Australia and New Zealand, including Lawrence (2003; 2007). The Australian Energy Regulator (AER) has indicated that MTFP analysis is one of the benchmarking methods it will use (together with stochastic frontier analysis and DEA) when benchmarking energy businesses as part of its regulatory price determination processes (AER 2013a p.13). For these reasons this study uses the MTFP method of index analysis. The Caves, Christensen and Diewert (CCD) multilateral translog index is given by: (4.1) log (TFP m /TFP n ) = i (R im +R i * ) (log Yim - log Y i * )/2 i (R in +R i * ) (log Yin - log Y i * )/2 j (S jm +S j * ) (log Xjm - log X j * )/2 + j (S jn +S j * ) (log Xjn - log X j * )/2 Where R i * (S j *) is the revenue (cost) share averaged over all utilities and time periods and log Y i * (log X j *) is the average of the log of output i (input j). In this analysis we have three outputs (the number of customers and the volumes of water supplied and sewage treated) and, 12 The transitivity property states that direct comparisons between observations m and n should be the same as indirect comparisons of m and n via any intermediate observation k. 'Characteristicity' refers to the degree to which weights are specific to the comparison at hand. 19

25 hence, i runs from 1 to 3. We have 2 inputs (non-capital and capital inputs) and, hence, j runs from 1 to 2. The Y i and X j terms are the output and input quantities, respectively. The R i and S j terms are the output and input weights, respectively. Formula (4.1) gives the proportional change in MTFP between two adjacent observations (denoted m and n). An index is formed by setting some observation (usually the first in the database) equal to one and then multiplying through by the proportional changes between all subsequent observations in the database to form a full set of indexes. The index for any observation then expresses its productivity level relative to the observation that was set equal to one. However, given the invariant nature of the comparisons, the result of a comparison between any two observations will be independent of which observation in the database was set equal to one. To calculate the MTFP index it is necessary to have output and input cost shares, which normally requires information on prices. In utility industries the billing units used in pricing do not necessarily coincide with the outputs, and the revenue shares associated with different outputs need not be closely related to the marginal costs of producing those outputs. In these circumstances it is desirable to use marginal costs in place of output prices to calculate the output weights (Fuss 1994). Lawrence (2007) uses cost shares associated with each output derived from an econometric cost function. For this study we have estimated a cost function for water businesses using the translog cost function specification. The elasticities of the cost function with respect to the outputs are used as weights for aggregating outputs (i.e. the R s in equation 4.1). Cost share weights are used for weighting inputs and these can be calculated using an exogenous or endogenous rate-of-return (Lawrence 2007). This study uses the exogenous rate-of-return approach to calculate the cost shares. The weight given to non-capital inputs is the ratio of opex to calculated total economic cost and the capital input weight is equal to one minus the opex share. 4.2 Frontier Analysis Frontier models are motivated by the idea that certain economic relationships describe the optimal level of one variable as a function of a set of explanatory variables, when profit or social welfare are maximised. For example, the production function defines the maximum amount of output that can be produced with a given set of inputs. The cost function refers to the minimum cost that can be achieved while producing a given amount of output, and given the set of input prices. For these types of relationships it makes sense to view optimum level of production or cost as an efficiency frontier which can be approximated but not exceeded by the best performing firms in the industry. Deviations from the efficiency frontier represent individual firm inefficiencies. Deviations from the efficiency frontier can be measured from either an input-oriented or output-oriented direction. The input-oriented measure refers to the degree to which inputs could be equiproportionately reduced while still being able to produce an unchanged set of outputs. Inefficiency is measured by the amount of this contraction relative to the input level required by a technically efficient firm to produce the same set of outputs. The output- 20

26 oriented measure refers to the extent to which outputs could be equiproportionately increased using an unchanged set of inputs. Here, inefficiency is measured by the amount of the output expansion relative to the output level that can be achieved by a technically efficient firm using the same set of inputs. The two most widely adopted frontier methods are: Stochastic frontier analysis, which is an econometric method of fitting the frontier function making use of a composite error term which includes the classical white noise disturbance and a one-sided disturbance which represents inefficiency. Data envelopment analysis (DEA), which is a linear programming technique for enclosing a set of data points within the tightest possible linear concave space (for production functions) or convex space (for cost functions). There is a wide range of modelling techniques within each of these categories, each with different strengths and weaknesses. Comprehensive information on these methods is available in (Bogetoft 2012; Coelli et al. 2005; Fried et al. 2008). This study uses stochastic frontier analysis in preference to DEA for consistency with ESC (2012a) for the reasons given by the ESC (2012) and Cunningham (2013). The main elements of the analytical approach to econometric estimation of TFP used in this study are: the input-oriented distance function the Malmquist index of TFP, and stochastic frontier analysis Distance Functions Measures of inefficiency based on deviations from the efficiency frontier are closely related to concept of output-oriented and input-oriented distance functions. A distance function relates technical efficiency to the mix of inputs and outputs and firm-specific technical inefficiency. The output-oriented distance function describes a firm s combination of an output and inputs in terms of the maximum set of outputs that can be produced with a given set of inputs (a multi-output extension of the production function concept), and a firmspecific inefficiency in each period. Similarly, the input-oriented distance function describes firm s mix of outputs and inputs in terms of the minimum set of inputs that can produce a the given set of outputs, and a measure of the firm s inefficiency relative to that minimum set of inputs (see: Coelli et al ch.3). The input distance function is analogous to a cost function. The input-oriented distance function used in this analysis is assumed to have the translog form, which is consistent with Coelli et al. (2003) and Saal et al. (2007). There are M outputs and K inputs, and y and x represent the logs of outputs and inputs respectively. The indicators i and t refer to firm i and period t. 21

27 (4.2) M ln D I(i,t) = u i,t = α 0 + β m y m(i,t) + 1 β! m=1 2 mn y m(i,t) y n(i,t) + α k x k(i,t) m=1 n=1 k=1 M M K K K + 1 α 2 kl x k(i,t ) x l (i,t ) + γ km x k(i,t ) y m(i,t ) + φ m y m(i,t ) t + δ k x k(i,t ) t k=1 l=1 +λ 1 t λ 11 t 2 + θ j z j (i,t ) J j =1 K M k=1 m=1 where the z s are variables relate to business conditions. The inverse of the input distance function is a measure of Farrell input based efficiency of the firm, and lnd I is therefore the technical inefficiency of the firm. More information about the distance function specification and decomposition is provided in appendix A. M m=1 K k= Malmquist TFP Index The Malmquist TFP index is defined in terms of input-oriented or output-oriented distance functions. In the case of the input oriented distance function, the corresponding measure of technical efficiency is: (4.3)! TE I (x t,y t ) = 1 D I (x t,y t ) And the Malmquist productivity index is defined as (Färe et al. 1998): (4.4) M s I = TE s (y I t+1,x t+1 ) TE s! I (y t,x t ) where s represents the date of the technology. Usually, M I is evaluated at both s = t and s = t+1, and the geometric average of the two measures is used. As shown in appendix A, the rate of change in the (generalised) Malmquist productivity index can also be expressed as the difference between the rates of change in indexes of inputs and outputs where the weights of those indexes are elasticities of the distance function with respect to outputs and inputs respectively. That is, the elasticities of the distance function with respect to outputs and inputs are used as weights in the indexes of total output and total input rather than the revenue and cost shares used in the Törnqvist index formula. One of the benefits of this approach is that it does not require price data to calculate. This is the method used for calculating the TFP index used here Stochastic Frontier & Random Effects Models Two econometric approaches were used in the 2012 ESC study: stochastic frontier analysis 22

28 and the random effects model. Stochastic frontier analysis is an econometric method for fitting a function to data which represents an upper or lower bound to the observations. 13 It is used for estimating minimum or maximum value functions. In the stochastic frontier approach, firm-specific technical inefficiency is measured against an estimated efficiency frontier. The most efficient firms are on the frontier, and the others have positive inefficiency. The efficiency frontier is subject to random disturbance. The firm-specific inefficiencies are assumed in the preferred set of estimates in this study to have a truncated-normal distribution. The random effects model is an econometric technique for analysing panel data (i.e. combined cross-sectional time series data) which includes a random cross-sectional disturbance (termed here the firm-specific effect ) as well as a random disturbance over all observations in the sample (i.e. cross-sectional & time series without distinction). In the random effects model, the estimated input distance function is a central or representative estimate, rather than a frontier. The firm-specific effects are assumed to be distributed symmetrically and normally. Although the firm-specific effects do not have the strict interpretation as measures of technical inefficiency, they can be interpreted as such. The random effects model is a useful comparative method, particularly when the estimated firm-specific inefficiencies appear to be close to normally distributed. If firms were observed to be more bunched toward an efficiency frontier, the stochastic frontier model would have clearer advantages. In this study we find the stochastic frontier and random effects models produce similar estimates overall. The econometric approaches have advantages over the index approach because they can take account of a range of interactions and some environmental or business conditions factors that the index approach cannot. 14 The econometric approaches also permit productivity change to be decomposed into its sources, including: technology improvement, reduction in the inefficiency of utilities, and returns-to-scale effects due to changes in the level of outputs. This is explained in appendix A. When estimating equation (4.2) using the stochastic frontier method, the stochastic disturbance used in the econometric model is decomposed into a white noise term (v it ) and a strictly positive inefficiency term (u it ), as shown in equation 4.5. The inefficiency term has a time-varying decay over the sample period at a common rate given by the parameter η ( eta ) and it has a cross-sectional component (u i ) which is positive and distributed according to a truncated normal distribution. The point of truncation is given by the parameter µ ( mu ) (see: Coelli et al. 2005). Eta and mu are reported with the other parameters in the regression output presented in section 5.2. (4.5)! ε = ν u ;!!!!!ν N(0,σ 2 ); it it it it ν! u = exp η(t T it { ) i }u i ;!!!!!u i N + (µ,σ 2 u ) 13 This frontier is subject to white noise random disturbance, hence the name stochastic frontier. 14 The business conditions factors included in the econometric models are discussed in section

29 The stochastic frontier method has the advantage that it can be used to estimate the coefficients of equation (4.2) while also providing estimates of the technical efficiency of each GDB in the sample. The estimated stochastic frontier models are presented in section 5.2 and the estimated random effects models are presented in appendix B. The results of applying these methodologies are discussed in the following section. 24

30 5 RESULTS The purpose of this section is to present the results of the Multilateral TFP index analysis and of the econometric stochastic frontier analysis. These results relate primarily to the trends in productivity and the comparative levels of technical efficiency of the water businesses included in the sample. 5.1 Index Analysis Multilateral TFP indexes were calculated for all utilities in the sample, and using two alternative measures of capital inputs: the deflated written down value of fixed assets, and the length of water supply and sewerage mains. The results presented in the remainder of this section represent the geometric average of the Multilateral TFP indexes calculated with these two capital measures. Key results of the Multilateral TFP index analysis are presented in Table 5.1. Over the whole sample the average rate of productivity decline between 2006 and 2013 was 1.9 per cent per year. Between 2006 and 2013, for the four major Victorian metropolitan water distributors as a group, TFP decreased on average at a rate of 1.1 per cent per year. Among them, the TFP of South East Water and Yarra Valley Water decreased at average rates of 0.5 and 0.4 per cent respectively over the same period. City West Water s TFP decreased at an average rate of 1.2 per cent and Barwon Water s TFP declined at an average rate of 2.4 per cent. Over the same period TFP declined on average in each of the other regions. The average TFP of the regional Victorian urban water utilities declined at an annual rate of 1.7 per cent on average. The TFP of major interstate water businesses declined at an average annual rate of 1.7 per cent per year, while the TFP of interstate regional water businesses declined at a rate of 2.1 per cent per year on average. Table 5.1 also shows measure of productivity levels for 2013 using multilateral TFP indexes. These measures do not control for economies of scale, and tend to be higher for larger water businesses. The productivity level of ACTEW 1998 is used as the base of this index (ie, equal to 1.0). In summary: The average productivity level of the four major Victorian water distributors was comparatively high at 1.5. Among them South East Water and Yarra Valley Water had the highest productivity levels, at 1.9 and 1.8 respectively. Barwon Water s productivity was the lowest of these four at 1.0. The average productivity levels in the other regions were: 0.9 for the regional Victorian water businesses; 1.2 for the major interstate water businesses; and 0.8 for the regional interstate water businesses. The average productivity level for the sample as a whole was 0.9 in 2013 compared to ACTEW

31 Table 5.1: Multilateral TFP indexes by utility type, MTFP growth MTFP index Year Barwon Water -2.43% City West Water -1.24% South East Water Ltd -0.47% Yarra Valley Water -0.41% Average Major Vic -1.14% Gippsland Water -5.65% Central Highlands Water -0.75% Coliban Water -0.98% Goulburn Valley Water -2.66% East Gippsland Water -2.81% GWMWater -0.70% Lower Murray Water 1.26% North East Water -0.18% South Gippsland Water -3.50% Wannon Water -0.97% Western Water -4.17% Westernport Water 0.10% Average Regional Vic -1.75% Average major interstate -1.67% Average regional interstate -2.09% All utilities -1.90% Source: Economic Insights calculations. Table 5.2 shows the Multilateral TFP indexes for each Victorian water distribution business over the periods 1998 to 2013 for larger utilities and 2006 to 2013 for smaller utilities. 26

32 Table 5.2: Multilateral TFP indexes by Victorian utility, (& for smaller utilities) Year Barwon Water City West Water South East Water Yarra Valley Water Gippsland Water Central Highlands Coliban Water Goulburn Valley Water Average Annual Change % 3.54% 2.33% 2.13% -4.44% -2.01% -0.72% 0.13% % -1.24% -0.47% -0.41% -5.65% -0.75% -0.98% -2.66% % 1.28% 1.01% 0.94% -5.05% -1.43% -0.84% -1.18% Source: Economic Insights calculations. 27

33 Table 5.2: (cont.) Year East Gippsland Water GWMWater Lower Murray Water North East Water South Gippsland Water Wannon Water Western Water Westernport Water Average Annual Change % -0.70% 1.26% -0.18% -3.50% -0.97% -2.78% 0.10% Source: Economic Insights calculations. 28

34 5.2 Stochastic Frontier Analysis The preferred econometric modelling approach in this study is the stochastic frontier analysis. This section presents the results from estimating the input-oriented distance function using this method. A summary of the results from estimating the random effects model is presented in appendix B Econometric results Table 5.3 presents the results of estimating two stochastic frontier models that are identical except that they use different measures of capital inputs. The first model uses the deflated written down value of fixed assets. The second uses the length of water supply and sewerage mains. The results presented in the remainder of this section were obtained using an average of these two models Productivity trends by utility type A summary of the estimated productivity trends, based on the stochastic frontier model, is plotted in figure 5.1. The same data is presented in table 5.4. Figure 5.1: Stochastic frontier: TFP indexes,

35 Table 5.3: Stochastic frontier regression estimates Capital = real WDV 1 Capital = Mains length 2 Coefficient Variable Estimate z statistic 3 Estimate z statistic α β 1 customers (y 1 ) β 2 water supply (y 2 ) β 3 sewerage services (y 3 ) α 1 capital inputs (x 1 ) β 11 y 1 y 1 / β 12 y 1 y β 13 y 1 y β 22 y 2 y 2 / β 23 y 2 y β 33 y 3 y 3 / α 11 x 1 x 1 / γ 11 x 1 y γ 12 x 1 y γ 13 x 1 y λ 1 t (year 1997) λ 11 t φ 1 y 1 t φ 2 y 2 t φ 3 y 3 t δ 1 x 1 t θ 1 trade-waste share θ 2 groundwater share θ 3 recycled share θ 4 sewerage penetration θ 5 net greenhouse (water) mu eta Log likelihood: Log likelihood: Critical z-statistics for testing are: 1.289, 1.658, and for the 20, 10, 5 and 1 per cent significance levels, respectively. A 5 per cent level of significance is used as the standard measure and less than 1 per cent is considered to be a very high level of significance. Results at the 10 per cent level of significance are also considered to be statistically meaningful. 30

36 Table 5.4: Stochastic frontier: TFP indexes by utility type, Year Major Vic Major interstate Regional Vic Regional interstate Total Sample Average Annual Change % 0.94% -1.08% 1.61% 0.77% % -0.95% -1.09% -1.30% -1.19% % 0.06% -1.08% 0.24% -0.15% Source: Economic Insights calculations. Some of the main findings of the stochastic frontier analysis are as follows: Over the period 1998 to 2006 the urban water industry in Australia had modest productivity improvement of approximately 0.8 per cent annually on average. However, from 2006 to 2013, productivity is estimated to have declined at an average rate of 1.2 per cent per year. Over the whole period from 1998 to 2013, there is estimated to have been a marginal productivity decline of 0.15 per cent per year. The major Victorian and regional interstate water businesses enjoyed a comparatively high rate of productivity growth over the period from 1998 to 2006, of 1.9 per cent and 1.6 per cent respectively. However, the regional Victorian water utilities had declining productivity, at an average rate of -1.1 per cent per year. From 2006 to 2013 the productivity of the four major Victorian water distribution businesses decreased incrementally at an average annual rate of 0.4 per cent. The productivity of the major interstate water utilities decreased an average rate of approximately 1.0 per cent per year. These findings are broadly consistent with those of the previous study. From 2006 to 2013 the productivity of Victoria s regional water businesses decreased an average rate of approximately 1.1 per cent per year. The interstate regional water businesses also had had declining productivity, at an average rate of -1.3 per cent per 31

37 year. The findings for the Victorian regional water businesses are broadly consistent with those of the previous study. However, the interstate regional water businesses are found to have a faster rate of productivity decline than previously, due to larger productivity decreases in the period from 2010 to Considering the whole period from 1998 to 2013, the major Victorian utilities averaged productivity growth of approximately 0.8 per cent per year. However, the average productivity of the regional Victorian water businesses is estimated to have declined at an average annual rate of 1.1 per cent over the same period. For the interstate major and regional water businesses, the level of productivity in 2013 was essentially unchanged from its level in By definition, the rate of productivity growth is the difference between the rate of output growth and the rate of input growth. Estimated output and input indexes and growth rates, again based on the stochastic frontier model, are presented in tables 5.5 and 5.6. Table 5.5: Stochastic frontier: Output indexes by utility type, Year Major Vic Major interstate Regional Vic Regional interstate Total Sample Average Annual Change % 2.18% 1.65% 3.05% 2.22% % 1.23% 1.46% 1.48% 1.46% % 1.73% 1.56% 2.31% 1.87% Source: Economic Insights calculations. A comparison of the figures in these two tables highlights an important finding. The average growth of output for water utilities in Australia as a whole did show some slowing, averaging 1.5 per cent per year in the period 2006 to 2013, compared to 2.2 per cent in the period 1998 to Most of this decline occurred interstate in both the major and especially the regional 32

38 areas. In Victoria, there was a minor decline in the rate of output growth for the major water distributors, but very little decline in the rate of output growth in regional areas. In contrast the rate of growth of inputs has increased in most regions over the period from 1998 to On average for all of the utilities in the sample, the quantity of inputs increased at an average annual rate of 2.7 per cent in the period 2006 to 2013, compared to a rate of 1.4 per cent in the period 1998 to The sharp increase in the rate of input growth in the 2006 to 2013 period is particularly striking, and not only exceeds the rate of output growth in the same period, but also exceeds the higher rate of output growth that occurred in 1998 to The rate of input growth of the major Victorian water distribution businesses was lower than that of the other regions over the whole period from 1998 to 2013, and this must be considered as the key factor in their comparatively better productivity performance. Table 5.6: Stochastic frontier: Input indexes by utility type, Year Major Vic Major interstate Regional Vic Regional interstate Total Sample Average Annual Change % 1.22% 2.76% 1.41% 1.44% % 2.19% 2.57% 2.82% 2.68% % 1.68% 2.67% 2.07% 2.02% Source: Economic Insights calculations. 33

39 5.2.3 Decomposition of productivity trends The method of decomposing productivity trends is detailed in Appendix D. The components of these trends include: Changes in technical efficiency: the movement of firms productivity towards or away from the efficiency frontier Technical change: shifting of the efficiency frontier over time Returns-to-scale effects: improvement in productivity as output increases due to economies of scale Business conditions effects: changes in characteristics of the market being served that are favourable or unfavourable to productivity improvement. A decomposition of the average productivity of all water businesses into these four effects is shown in Table 5.7 for the period 1998 to The main findings of this decomposition are as follows: Changes in technical efficiency appear to have had a relatively steady negative effect on productivity of between 0.7 and 0.8 percentage points per year. This suggests that most firms included in the study have been moving further away from best practice. Technical change is estimated to have been positive over the period from 1998 to 2006, improving at a rate of approximately 1.3 per cent per year. However, technical change has been negative in the period from 2006 to 2013, decreasing at a rate of 0.7 per cent per year. Negative technical change is possible if external factors, such as changes in technical regulation, increase the amount of inputs that a best practice business would require to produce a given set of outputs. Over the whole period from 1998 to 2013, technical change is estimated to have contributed 0.4 percentage points per year to productivity growth. Changes in business conditions did not have any significant effect on productivity growth. Returns-to-scale effects had a small positive effect on productivity, estimated at approximately 0.3 percentage points per year. Table 5.8 shows the decomposition of the average productivity of Victorian water businesses over the period 2006 to Changes in technical efficiency had negative effect on productivity of approximately 0.8 percentage points per year, and technical change was slightly negative over this period, both consistent with the national pattern previously discussed. Returns-to-scale effects were positive, contributing approximately 0.35 percentage points per year, while changes in business conditions did not have any significant influence. The average rate of productivity growth among Victorian water businesses of -0.9 per cent per year over the 2006 to 2013 was only slightly better than the national average. 34

40 Table 5.7: Stochastic Frontier: TFP decomposition, Australia Year Technical Efficiency Technical Change Business Conditions Returns-to- Scale Effect Average Annual Change % 1.30% -0.09% 0.30% 0.77% % -0.69% -0.06% 0.37% -1.19% % 0.37% -0.07% 0.34% -0.15% Source: Economic Insights calculations. TFP Table 5.8: Stochastic Frontier: TFP decomposition, Victoria Year Technical Efficiency Technical Change Business Conditions Returns-to- Scale Effect Average Annual Change % -0.47% 0.00% 0.35% -0.92% Source: Economic Insights calculations. TFP 5.3 Comparative trends for Victorian water businesses Productivity growth outcomes for individual Victorian water businesses over the period from 2006 to 2013 are shown in Table 5.9. There was some variance in the productivity growth of 35

41 the four major utilities, the highest being Yarra Valley Water with 0.3 per cent productivity growth per year, and the lowest being Barwon Water with a decline of 1.3 per cent per year. Nine of the twelve regional water businesses had negative productivity growth during this seven year period (not including Westernport Water, which had essentially unchanged productivity). Most of the utilities with the largest rates of productivity decline had high rates of growth in inputs. Six of the sixteen Victorian water businesses included in this study are estimated to have increased the quantity of inputs at an average rate exceeding three per cent per year during the period 2006 to These included Barwon Water, City West Water, Gippsland Water, East Gippsland Water, South Gippsland Water and Western Water. Table 5.9 Stochastic frontier: Output, input & TFP growth by utility, Utility Output growth p.a. Input growth p.a. TFP growth p.a. Barwon Water 1.87% 3.20% -1.29% City West Water 2.73% 3.02% -0.29% South East Water Ltd 1.26% 1.67% -0.40% Yarra Valley Water 1.68% 1.33% 0.34% Central Gippsland Water 1.76% 4.55% -2.68% Central Highlands Water 1.84% 2.56% -0.71% Coliban Water 1.29% 2.17% -0.86% Goulburn Valley Water 0.10% 1.74% -1.61% East Gippsland Water 1.49% 3.51% -1.95% GWMWater 0.25% 0.92% -0.66% Lower Murray Water 1.72% 0.49% 1.23% North East Water 1.50% 1.38% 0.11% South Gippsland Water 1.51% 4.14% -2.53% Wannon Water 0.94% 1.55% -0.60% Western Water 2.66% 5.47% -2.67% Westernport Water 2.46% 2.50% -0.04% Total 1.56% 2.50% -0.92% Source: Economic Insights calculations. 5.4 Comparative Technical Efficiency The stochastic frontier model provides estimates of the technical efficiency of each utility relative to the efficient frontier. We have divided these efficiency scores by the average for all of the Australian water businesses, to provide a measure of relative efficiency. Table 5.10 and Figure 5.2 present the relative technical efficiency scores estimated for all businesses in the sample. Figure 5.2 also includes the statistical confidence interval relating to those estimates. Figure 5.3 presents similar information for only the Victorian water businesses. Figure 5.4 presents the same information, but including only the Australia s major water businesses. Figure 5.5 presents the comparative technical efficiencies of all of the regional water businesses in the sample, both in Victoria and interstate. 36

42 These technical inefficiency measures need to be interpreted cautiously. A number of business conditions variables have been included in the econometric analysis to control for some of the characteristics of the distinct markets in which urban water utilities operate. However, it is likely there will be other potentially significant exogenous factors that produce heterogeneity, such as topography, climate, customer density or regional input cost differences. Relevant determinants may not be measured, or may be unknown, and thus cannot be included in the analysis. This is the general problem of unobserved firm-specific heterogeneity. An assumption that all residual firm-specific effects are entirely due to technical efficiency would be incorrect (Cunningham 2012 p.8). Here we will limit our focus on the quartile of technical efficiency within which a business is estimated to belong. Indicatively, the upper quartile of the range of relative technical efficiencies includes those greater than 1.1, and the lowest quartile encompasses those below Victorian water businesses fell within these quartiles as follows: Victorian water utilities estimated to be in the highest quartile of technical efficiency include Westernport Water, South East Water, and Lower Murray Water. Yarra Valley Water and South Gippsland Water are estimated to be in the second quartile. Those estimated to be in the third quartile include: GWMWater, City West Water, East Gippsland Water, Western Water, and Goulburn Valley Water. Those estimated to be in the lowest quartile of technical efficiency include: Wannon Water, Gippsland Water, Barwon Water, Central Highlands Water, Coliban Water and North East Water. In this study the four major urban water distribution businesses, taken as a group, were found to be as efficient as the average water business. The major urban water businesses interstate were found to be, on average, approximately 5 per cent less efficient than the average water business. Regional Victorian water businesses were found to be, on average 4 per cent less efficient than the average water business. The regional water distribution businesses interstate were found to be approximately 2 per cent above the average efficiency of water businesses. These findings are to some extent similar to the ESC s 2012 study, although generally the degree of the differences in average technical efficiency between these groups of utilities appears to be much less pronounced. In part this may reflect the wider sample used in the present study. 37

43 Table 5.10 Stochastic Frontier: Relative technical efficiency estimates (sample average = 1.0) Utility TE Utility TE Utility TE ACTEW Kempsey Shire Council South East Water Ltd Hunter Water Corporation Lismore City Council Yarra Valley Water Sydney Water Corporation Orange City Council Central Gippsland Water Gosford City Council Queanbeyan City Council Central Highlands Water Wyong Shire Council Tamworth Regional Council Coliban Water Albury City Council Wingecarribee Shire Council Goulburn Valley Water Coffs Harbour City Council Power and Water Darwin East Gippsland Water MidCoast Water Power and Water Alice Springs GWMWater Port Macquarie Hastings Council Brisbane Water Lower Murray Water Shoalhaven City Council Gold Coast City Council North East Water Tweed Shire Council Cairns Water and Waste South Gippsland Water Wagga Wagga / Riverina Water Fitzroy River Water Wannon Water Ballina Shire Council Queensland Urban Utilities Western Water Bathurst Regional Council Unitywater Westernport Water Bega Valley Shire Council Water and Waste Services Mackay Water Corporation Perth Byron Shire Council Wide Bay Water Water Corporation - Mandurah Clarence Valley Council Ipswich Water Water Corporation - Albany Essential Energy Logan Water Aqwest & Water Corp. Bunbury Dubbo City Council SA Water - Adelaide Water Corporation - Geraldton Eurobodalla Shire Council Barwon Water Kalgoorlie-Boulder & Water Corp Goulburn Mulwaree Council City West Water

44 Figure 5.2: Stochastic frontier: Comparative technical efficiency levels, Australia

45 Figure 5.3: Stochastic frontier: Comparative technical efficiency levels, Victoria

46 Figure 5.4: Stochastic frontier: Comparative technical efficiency levels, Major utilities Australia

47 Figure 5.5: Stochastic frontier: Comparative technical efficiency levels, Regional utilities Australia