i
THE PRICE ELASTICITY OF WATER DEMAND FOR SYDNEY RESIDENTIAL
CONSUMER SEGMENTS – A PANEL ECONOMETRICS STUDY.
by
Santharaj...
ii
ORIGINALITY STATEMENT
‘I hereby declare that this submission is my own work and to the best of my knowledge it
contains...
iii
ACKNOWLEDGEMENTS
I would like, first and foremost, to express my sincere gratitude to my principal supervisor, Dr Mari...
iv
ABSTRACT
Planning for future water demand and price paths requires a nuanced understanding of the impact of
potential p...
v
T a b l e o f C o n t e n t s
Table of Contents............................................................................
vi
3.2.7 Household information...............................................................................................
vii
6.3.7 Segment 7 – Renters (Department of Housing-DoH)....................................................... 66 
6.3.8...
viii
L i s t o f T a b l e s
Table 2-1: Water rates readings for an example property for the period 2007-2008................
ix
L i s t o f F i g u r e s
Figure 2-1: Example average daily consumption between the meter read dates......................
x
G l o s s a r y
BASIX. Building Sustainability Index is a scheme introduced by the government of New South Wales,
Austra...
xi
1
C h a p t e r 1 - I N T R O D U C T I O N
1.0 Introduction
Despite the Earth appearing to have an abundance of water, it...
2
1.1 Purpose and Objectives of the Study
Planning for future water demand and price paths requires a nuanced understandin...
3
Previous studies such as Worthington and Hoffman (2008) have identified that differences in
household income and the nat...
4
1.2 Structure of the thesis
There are seven (7) chapters in this thesis. Chapter 2 presents the literature associated wi...
5
C h a p t e r 2 - L I T E R A T U R E R E V I E W
2.0 Introduction
Accurate measure of the price elasticity of demand fo...
6
types of water consumption, by different types of dwellings, as well as leaks from water pipes. As for
monthly consumpti...
7
Figure 2-1: Example average daily consumption between the meter read dates
Source: Based on Sydney Water data sets
By mu...
8
The method so far discussed to calculate the average consumption of water induces a measurement
error in the dependent v...
9
between the years 1991 and 1994 where increasing block tariffs occurred, although minimal. Thus the
period between June ...
10
blocks of temperature ranges, namely (i) 4-21 C and (ii) above 21 C, where water usage does not
change much.
Bamezai (1...
11
indicate the proportion of the population over sixty-four years and under nineteen years. Warner (1996)
did not include...
12
luxury goods (e.g., swimming pools and in-ground garden irrigation systems.) associated with
household water consumptio...
13
As water is an essential commodity, the consumer demand theory indicates that a minimum amount of
water demand is expec...
14
The household size is arguably the most important factor in modelling water demand. However, it is
not easy to obtain i...
15
C h a p t e r 3 - T H E O R E T I C A L F R A M E W O R K
3.0 Introduction
This chapter discusses the development of th...
16
3.1 Demand function
A general dynamic model specification for this study is expressed as,
)1.3(
springwinterautumnsumme...
17
3.2 Water demand and explanatory variables
It is impossible to capture all factors affecting household water consumptio...
18
Figure 3-1: Water usage spilt in 2009-10 between different type of users
Source: Derived from Sydney Water data sets
As...
19
Figure 3-2: Industrial usage and leakage over time
Source: Based on Sydney Water data sets
Thirdly, it may not fully ac...
20
Finally, as discussed earlier, artificially created ‘monthly consumption’ is not suitable for this study
either. Due to...
21
Figure 3-4: Service charge and usage charge distribution for a household uses 240kL per year
(in real terms)
Source: Co...
22
price for residential consumption is applied if the water consumption level is above 400 per kL per
household per year....
23
This particular observation includes 68% of July, 100% of August, 100% of September and 73% of
October.
Similarly, it i...
24
The final weather variable specification for each segment was highly dependent on the diagnostic test
results.
The Bure...
25
3.2.5 Government rules and regulations
Sydney water internal studies indicate that government rules and regulations reg...
26
couples without kids tend to move to a bigger house when they have children. Older parents tend to
move out of their la...
27
having a twenty-minute shower is likely to continue their behaviour regardless of change in water
price, if it is small...
28
3.3 Functional form of demand
The linear water demand functions are the easiest to use, as no transformation is require...
29
such direct concept of a choke price. Given the difficulties associated with measuring income at the
household level, t...
30
C h a p t e r 4 - M E T H O D O L O G Y
4.0 Introduction
The principal method of analysis applies techniques in econome...
31
As for Sydney Water, this is a starting point to see how different consumer segments are reacting to the
change in wate...
32
Figure 4-1: Socio-Economic-Demographic Factors
From Figure 4-1, the base category is identified to have the following c...
33
4.2 Study area and Study period
Sydney Water covers more than 12,000 square kilometres of area, servicing more than fou...
34
The Sydney Water billing database has approximately 1.7 million residential dwellings and 167
thousand non-residential ...
35
Despite of all these restrictions, the Sydney’s real water usage price (adjusted to inflation) did not
change much for ...
36
To address these issues, the study explored panel data analysis method. The panel data analysis has two
major specifica...
37
and Nosvelli (2007) used GMM on Italian municipality water data. On the other hand, Martinez-
Espineira (2007) used the...
38
C h a p t e r 5 – D A T A D E S C R I P T I O N
5.0 Introduction
The data used for this study mostly comes from Sydney ...
39
As Figure 5.1 shows, the compilation of various information is a complex process that requires
diligence and through kn...
40
5.2 Selection of Properties
It is best to have a ‘strongly balanced’ data set for a panel study. Hence, the following r...
41
Table 5-1: Descriptive statistics for all segments
Variable
Name
Description Min. Max. Mean Std.
Dev.
Unit
Subject A su...
42
0.38 kL (almost 60% of the mean), which also shows that there is a huge variation in consumption.
This was expected, as...
43
5.3 Consumer segments
As pointed out earlier, 28,473 properties were selected for the analysis. Table 5-2 shows the exa...
44
Figure 5-2: Daily average water consumption within study period for each consumer segment
Following from Figure 5-2, th...
45
Chapter 6 -M O D E L R E S U L T S A N D D I S C U S S I O N S
6.0 Introduction
Fourteen separate models were devised r...
46
As argued in Section 4.3 (Estimation Methods), the dynamic nature of model specification is likely to
produce biased es...
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
Price elasticity of consumer segments - Panel econometric study
of 124

Price elasticity of consumer segments - Panel econometric study

Published on: Mar 4, 2016
Source: www.slideshare.net


Transcripts - Price elasticity of consumer segments - Panel econometric study

  • 1. i THE PRICE ELASTICITY OF WATER DEMAND FOR SYDNEY RESIDENTIAL CONSUMER SEGMENTS – A PANEL ECONOMETRICS STUDY. by Santharajah Kumaradevan (Student ID: 17145653) A thesis submitted as part of the requirements for the degree of Master of Commerce (Honours) University of Western Sydney February 2013
  • 2. ii ORIGINALITY STATEMENT ‘I hereby declare that this submission is my own work and to the best of my knowledge it contains no materials previously published or written by another person, or substantial proportions of material which have been accepted for the award of any other degree or diploma at UWS or any other educational institution, except where due acknowledgement is made in the thesis. Any contribution made to the research by others, with whom I have worked at UWS or elsewhere, is explicitly acknowledged in the thesis. I also declare that the intellectual content of this thesis is the product of my own work, except to the extent that assistance from others in the project's design and conception or in style, presentation and linguistic expression is acknowledged.’ Signed …………………………………………….............. Date …………………………………………….............. This thesis is submitted to the University of Western Sydney in partial fulfilment of the requirement for the Master of Commerce (Honours) with Course Code 8003.
  • 3. iii ACKNOWLEDGEMENTS I would like, first and foremost, to express my sincere gratitude to my principal supervisor, Dr Maria Estela Varua, for her continuous support of my research, her patience, motivation, enthusiasm and understanding. I greatly appreciate her guidance during the research process and in the writing of this thesis. She would be my first point of contact if I ever undertake PhD research. I would also like to thank my co-supervisor, Dr Girijasankar Mallik, for his encouragement, insightful comments and constructive criticism. The topic of my thesis arose from research work that was being undertaken by Sydney Water. I have worked with many people who have made contributions to this research. I would like to express my gratitude to my work colleagues for their contribution, particularly to Frank Spaninks for sharing his in-depth knowledge of Sydney Water’s datasets. This study would not have been possible without the support of Sydney Water, especially in providing data and allowing me to undertake this study as part of my role. I’m extremely grateful for the support and guidance that I have received from my managers at Sydney Water: Barry Abrams, Lucinda Maunsell and Peter Cox. They have provided valuable input in the preparation and completion of this research. Thank you again, Barry Abrams, for supporting me in achieving my dream of completing this research, patiently going through modelling results and providing valuable direction. Last but not least, I would like to express my utmost gratitude to my wife, Apirami Kumaradevan, who has been my inspiration as we conquered the obstacles to completing this research. She performed multiple roles, such as wife, friend, colleague, teacher, boss and editor, to make this dissertation possible.
  • 4. iv ABSTRACT Planning for future water demand and price paths requires a nuanced understanding of the impact of potential price signals on customers' level of usage. Most existing water price elasticity studies examine the ability of water usage prices to influence aggregate water use. Unlike most studies, this study primarily aims to investigate the water price elasticity of different Sydney households, which are uniquely segmented, based on dwelling type, socio-economic status and geographical location. Separate dynamic panel data models were specified for each of the 14 segments, using the econometric estimation method of General Method of Momentum (GMM). A total of 18,892 households’ water consumption levels were individually recorded, between the period January 2004 and June 2010. The analysis found that not all households react to the change in water price in the same way. Households with relatively high discretionary water use were able to reduce their water use as the water price increased, whereas financially disadvantaged households did not show the same ability to change their water usage to compensate for the increase in water price. Policy makers are therefore encouraged to keep these differences in mind when they set the water price or consider using water price as a demand management tool during drought. In addition, the results further illustrate the need for any water consumption modelling to consider the different consumer segments and to refine the model specifications for each segment to provide accurate information relevant to the consumer segment of interest.
  • 5. v T a b l e o f C o n t e n t s Table of Contents........................................................................................................................................v  List of Tables..........................................................................................................................................viii  List of Figures........................................................................................................................................... ix  Glossary ......................................................................................................................................................x  Chapter 1-INTRODUCTION ....................................................................................................................1  1.0 Introduction ....................................................................................................................................1  1.1 Purpose and Objectives of the Study.............................................................................................2  1.2 Structure of the thesis.....................................................................................................................4  Chapter 2-LITERATURE REVIEW.........................................................................................................5  2.0 Introduction ....................................................................................................................................5  2.1 Measurement of water demand......................................................................................................5  2.2 Determinants of water demand......................................................................................................8  2.2.1 Price.......................................................................................................................................8  2.2.2 Weather Variables and Seasonality......................................................................................9  2.2.3 Household information...................................................................................................... 10  2.2.4 Property specific information............................................................................................ 11  2.3 Functional forms ......................................................................................................................... 12  2.4 Summary...................................................................................................................................... 13  Chapter 3-THEORETICAL FRAMEWORK ........................................................................................ 15  3.0 Introduction ................................................................................................................................. 15  3.1 Demand function......................................................................................................................... 16  3.2 Water demand and explanatory variables................................................................................... 17  3.2.1 Water demand defined....................................................................................................... 17  3.2.2 Price of water ..................................................................................................................... 20  3.2.3 Seasonality ......................................................................................................................... 22  3.2.4 Weather conditions ............................................................................................................ 23  3.2.5 Government rules and regulations..................................................................................... 25  3.2.6 Property size and garden size ............................................................................................ 25
  • 6. vi 3.2.7 Household information...................................................................................................... 25  3.2.8 Adaptation of water efficient appliances or activities....................................................... 26  3.2.9 Habits and culture .............................................................................................................. 26  3.2.10 Other water uses............................................................................................................... 27  3.3 Functional form of demand......................................................................................................... 28  Chapter 4 - METHODOLOGY .............................................................................................................. 30  4.0 Introduction ................................................................................................................................. 30  4.1 Residential consumer segments.................................................................................................. 30  4.2 Study area and Study period ....................................................................................................... 33  4.3 Estimation methods..................................................................................................................... 35  Chapter 5 –DATA DESCRIPTION........................................................................................................ 38  5.0 Introduction ................................................................................................................................. 38  5.1 Data sources................................................................................................................................. 38  5.2 Selection of Properties ................................................................................................................ 40  5.3 Consumer segments .................................................................................................................... 43  Chapter 6 -MODEL RESULTS AND DISCUSSIONS......................................................................... 45  6.0 Introduction ................................................................................................................................. 45  6.1 Step 1 – Experimentation with estimators.................................................................................. 47  6.1.1 Ordinary Least Square Estimator (OLS)........................................................................... 49  6.1.2 Fixed Effects Estimator (FE Model)................................................................................. 50  6.1.3 Random Effects Estimator (RE Model)............................................................................ 51  6.1.4 Fixed Effects Estimator with time-specific effects........................................................... 51  6.1.5 Initial GMM Estimator (Initial)......................................................................................... 52  6.2 Step 2 – Experimentation with GMM specification .................................................................. 53  6.2.1 Improved GMM Estimator (Improved) ............................................................................ 54  6.2.2 GMM Estimator (Final)..................................................................................................... 57  6.3 Consumer Segment models ........................................................................................................ 59  6.3.1 Segment 1 – Average Houses (Base Segment)................................................................. 63  6.3.2 Segment 2 – Townhouses.................................................................................................. 64  6.3.3 Segment 3 – Gardeners...................................................................................................... 64  6.3.4 Segment 4 – Pensioners..................................................................................................... 65  6.3.5 Segment 5 – Swimming Pools........................................................................................... 65  6.3.6 Segment 6 – Financially in trouble.................................................................................... 66
  • 7. vii 6.3.7 Segment 7 – Renters (Department of Housing-DoH)....................................................... 66  6.3.8 Segment 8 – Payment Plan................................................................................................ 67  6.3.9 Segment 9 – Cultural and linguistically diverse (CALD) ................................................ 67  6.3.10 Segment 10 & 11 – Low Income & High Income.......................................................... 67  6.3.11 Segment 12 – Water Conscious....................................................................................... 68  6.3.12 Segment 13 & 14 – Small lot size & large lot size......................................................... 68  6.4 Key problems faced..................................................................................................................... 69  6.4.1 Issue 1: Autocorrelation..................................................................................................... 69  6.4.2 Issue 2: Use of appropriate instruments............................................................................ 71  6.4.3 Issue 3: Aligning variables which are read in different intervals..................................... 72  Chapter 7 – CONCLUSIONS AND RECOMMENDATIONS............................................................ 74  REFERENCES........................................................................................................................................ 78  Appendix 1: Stata code and results for the base segment ...................................................................... 81  Appendix 2: How to obtain the standard errors of the long-run elasticities.......................................... 89  Appendix 3: Stata code and results for all segments.............................................................................. 92  Appendix 4: Interpretation of dummy variable in semi-log regression............................................... 112  Appendix 5: Consumer Segments......................................................................................................... 113
  • 8. viii L i s t o f T a b l e s Table 2-1: Water rates readings for an example property for the period 2007-2008...............................6  Table 3-1: Sydney water price structure 2010-2011 .............................................................................. 20  Table 5-1: Descriptive statistics for all segments................................................................................... 41  Table 5-2: Consumer segments and number of properties..................................................................... 43  Table 6-1: Summary of Results of Popular Methods of Estimation...................................................... 48  Table 6-2: Results from the improved GMM estimation for equation 6.2............................................ 55  Table 6-3: Results from the final model specification ........................................................................... 58  Table 6-4: Parameter estimates for all consumer segments................................................................... 61  Table 6-5: Long-run estimates and model diagnostics for all segments................................................ 62  Table 7-1: Comparing price elasticity estimates with previous Sydney based studies......................... 76
  • 9. ix L i s t o f F i g u r e s Figure 2-1: Example average daily consumption between the meter read dates......................................7  Figure 3-1: Water usage spilt in 2009-10 between different type of users............................................ 18  Figure 3-2: Industrial usage and leakage over time................................................................................ 19  Figure 3-3: Impact of water efficiency programmes.............................................................................. 19  Figure 3-4: Service charge and usage charge distribution for a household uses 240kL per year ........ 21  Figure 3-5: Inflation adjusted real price of fixed charges and usage charges over time....................... 21  Figure 3-6: Map of Sydney weather stations used for this study........................................................... 24  Figure 4-1: Socio-Economic-Demographic Factors............................................................................... 32  Figure 4-2: Sydney Water area of operations......................................................................................... 33  Figure 4-3: Timeline (bulk demand, real price increases, water conservations and levels of water)... 34  Figure 5-1: Data Sources......................................................................................................................... 38  Figure 5-2: Daily average water consumption within study period for each consumer segment......... 44  Figure 6-1: Flow chart diagram mapping the steps taken to reach the final consumer segment models45  Figure 6-2: Long-run price elasticity estimates with 90% confidence interval..................................... 63
  • 10. x G l o s s a r y BASIX. Building Sustainability Index is a scheme introduced by the government of New South Wales, Australia in 2004 to regulate the energy efficiency of residential buildings. Bulk water. Total water inputted into a water distribution network. This includes both metered and unmetered water. Choke price. The price at which demand for an item shifts to zero. Drought restrictions. Mandatory restrictions on the type, method and time people can undertake outdoor water use. Gigalitre. Equal to one thousand mega litres. Kilolitre. Equal to one thousand litres. Long-run marginal cost. The expected cost of bringing forward an extra unit of supply in the long term, including associated capital expenditure for infrastructure. Mega litre. Equal to one thousand kilolitres. Price elasticity of demand. The responsiveness of demand to changes in price. This is calculated as the percentage change in demand due, divided by the percentage change in price. Scarcity pricing. A charge that reflects the value of water in alternative uses, during periods of water shortages. Short-run marginal cost. The cost of providing an additional unit of supply to meet demand in the short term, reflecting the highest value use to which a commodity can be put in periods of shortage. Water conservation activities. Includes leak reduction, water efficiency programs and recycling. WELS. Water Efficiency Labelling and Standards is water efficiency labelling scheme that requires certain products to be registered and labelled with their water efficiency in accordance with the standard set under the national Water Efficiency Labelling and Standards Act 2005.
  • 11. xi
  • 12. 1 C h a p t e r 1 - I N T R O D U C T I O N 1.0 Introduction Despite the Earth appearing to have an abundance of water, it is estimated that only around 0.6% of the Earth's water is actually fresh and usable by the population (Michael et al., 2007). While the recent rains in New South Wales (NSW) have temporarily eased the drought issue, the underlying stresses remain. Records show that since 1970, only the years 1974, 1984, 1989 and 2000 have been considered to be drought free in NSW. Sydney, the state capital of NSW, is the most populous city in Australia. It is also labelled as Australia’s ‘economic powerhouse’ and is voted as one of the world's top most liveable cities but drought, aging assets and rapid population growth are some of the key challenges it faces when it comes to water provision. Sydney Water, the state agency responsible for the provision of water, first developed a Water Conservation Strategy in 1995. Sydney Water supplies wastewater, recycled water and some stormwater services to over 4.6 million people in Sydney, the Illawarra and the Blue Mountains. The strategy was able to reduce the demand for water by over 100 billion litres of water each year. However, this strategy and the underlying programs have now reached maturity and are no longer commercially viable due to a decrease in the adoption of these programs. Also Sydney Water’s focus for water efficiency has changed taking into account the increased cost of living. Most of the voluntary water efficiency programs ceased last 30 June 2011. Further, the majority of these water efficiencies are achieved with regulations such as the Building Sustainability Index (BASIX) and Water Efficiency Labelling and Standards (WELS). Furthermore, based on Sydney Water’s previous experiences it may prove difficult to implement water efficiency programs within a short period of time to manage droughts that may occur in the future. Given these pressures, there is a considerable interest in the role that water usage prices can play in managing the supply and demand for water. Barker et al. (2010) argued that scarcity pricing is a more preferred option to imposing restrictions on water use in managing drought. Furthermore, few studies such as Hewitt and Hanemann (1995) produced higher price elasticity estimates near -1.6 which shows that any policy change such as changes in the price of water should be thoroughly studied and evaluated as it affects the general public.
  • 13. 2 1.1 Purpose and Objectives of the Study Planning for future water demand and price paths requires a nuanced understanding of the impact of potential price signals to consumers' level of usage. An analysis of consumer behaviour over different community segments may allow more subtle variations to be found regarding water use. In Australia, water usage prices have considerably increased nationwide in recent years. In assessing the role of water usage prices, the two key considerations for stakeholders (e.g., utilities, regulators and policy makers) are:  the aggregate impact on water use from changes in water usage prices, and  the distribution of the impact in water usage charges across the community. Although there are a number of studies such as Grafton et. al (2007) that have been undertaken on the price elasticity of demand of water in Sydney, most of these studies examine the ability of water usage prices to influence aggregate water use. A recent study by Grafton and Ward (2007) sought to measure the impact of price changes in Sydney’s bulk water use. Unfortunately, due to the unavailability of disaggregated data, none of these studies were able to access the distribution of price increases across the community. The lack of access to this type of data can be partly attributed to privacy issues. It is almost impossible for any researcher outside of Sydney Water Corporation to individually identify residential segments and estimate price elasticity of demand for water from these segments. On the other hand, water utilities may also have difficulty finding an internal employee who has the necessary skills and interest to assemble all the complex datasets and have the time and ability to analyse these data using sophisticated econometrics techniques. Further, it can be difficult to create a rationale for the organisation such as Sydney Water to allow such a researcher to spend more than a year to undertake this type of study. An understanding of the distribution of the impact in the change of water usage charges across the community is important, as it provides insight into potential social issues, such as the financial stress experienced by households due to their inability to pay their utility bills. Given the importance of water in every household, this study primarily aims to investigate the changes in water demand by the different segments of Sydney households, as a result of a change in the price of water.
  • 14. 3 Previous studies such as Worthington and Hoffman (2008) have identified that differences in household income and the nature of property ownership may influence the change in demand. Often times it is argued that the differences are not captured or are unobserved by researchers. To address these concerns, this study segments the households using a combination of household characteristics. Specifically, this study will identify various segments based on the following combinations of household characteristics:  Type of dwelling (e.g., building structure, property ownership, land size)  Economic status (e.g., wealthier suburbs vs. poorer suburbs)  Financial status (e.g., income levels)  Geographical location (e.g., coastal areas vs. inland areas)  Social status (e.g., pensioners) Using these characteristics, fourteen segments of Sydney households were identified and is used in this study. This method of segmentation is done to reduce the heterogeneity among households. A separate dynamic panel data model specification to each of the fourteen segments using an econometric estimation method known as the Generalised Method of Moments (GMM) is applied in this study. The GMM is specified to estimate the residential price elasticity of water demand for each segment. On average, each segment consists of around 1300 individual households. The recorded water consumption level for each of these households, for the period between January 2004 and June 2010, were used. The price elasticity estimates were then utilised to examine the trends within each market segment. Specifically, this thesis hopes to answer questions such as:  Can scarcity pricing be used to manage water demand in any particular consumer segment?  Do consumer segments need to be considered when water prices are set?  Do socially disadvantaged consumer segments appear less or more able to respond to changes in water usage prices?  Does the current water price structure enable substantial water savings in any particular consumer segment?
  • 15. 4 1.2 Structure of the thesis There are seven (7) chapters in this thesis. Chapter 2 presents the literature associated with this field of study. This chapter is divided into three main sections based on the method of modelling price elasticity of demand, measurement of the dependent variable, selection of independent variables and specification of functional forms. Each section reviews studies undertaken in Sydney, in Australia, or in other countries. In Chapter 3, the theoretical framework chosen for the study is discussed. This is followed by the exploration of the methodology used, in Chapter 4. Chapter 4 discusses how each consumer segment is created and the characteristics of each. For this study, the first segment is designated as the base segment. Households classified in this segment have the following physiognomies.  Owner-occupied houses with median property lot size  No participation in any water efficiency programmes  Located in the middle ring of suburbs  Classified as a financially average household The base segment is selected to represent the average middle class household in Sydney. The houses generally have more than two bedrooms, and have gardens. Families with children tend to largely occupy these houses. The rest of the households were then classified into other segments, as discussed in Chapter 4. In addition, the study area, estimation and choice of time period are described in this chapter. Chapter 5 presents the data and outlines how households were selected for the analysis. The results of the base modelling are discussed in Chapter 6. This modelling is on owner-occupied single-dwellings, and water consumption observed between July 2004 and June 2009. Findings from the base model were utilised to construct a model for each consumer segment. The modelling results for each consumer segment are analysed and presented in Chapter 6. Chapter 6.4 identifies the key problems and limitations of the study and finally, Chapter 7 presents the conclusions.
  • 16. 5 C h a p t e r 2 - L I T E R A T U R E R E V I E W 2.0 Introduction Accurate measure of the price elasticity of demand for potable water is critical in understanding and forecasting the effect of changes in price of water on demand and supply, on securing water availability for future generations and on revenue. Further, it is also important to identify the contributing factors that influence water consumption in order to formulate well-informed policy decisions. To date there are only two studies that were carried out for Sydney that determines the price elasticity of demand for water using household data. Abrams et al. (2012) attempted on a larger scale on the price elasticity of demand for water study at the household level. The primary aim of Abrams et al. (2012) was to calculate the price elasticity of water demand for water using the first difference model. They assumed that some of the time invariant factors such as household size and lot size were constant and were therefore not included in their model specification. The succeeding sections in this chapter will present previous studies that were useful in the development and in the analysis of the model proposed for this. Chapter 3 will then discuss the theoretical framework. 2.1 Measurement of water demand Warner (1996) used yearly bulked water consumption data from 1960 to 1994. To use yearly aggregated bulk water data to ascertain the impact of a change in price of water has limitations as this method may not fully capture the variability of the data. Grafton and Kompas (2007), on the other hand, used daily bulk demand data for water. Although it is easier to capture variability in water demand due to factors such as changes in weather conditions, it still has all the problems associated with the use of bulk data, as discussed further in Section 3.2.1. In the study by Worthington and Hoffman (2008), they concluded that the aggregation of households produces fairly similar results to those that use daily data. However, it is important to note that Worthington and Hoffman (2008) are not advocating the aggregation of data across industries, but rather are reporting the results generated from individual residential data and aggregated residential water usage. Review of existing literature shows that most of the existing Sydney based studies used bulk water data or monthly metered data. Bulk water data introduces aggregation bias due to the inclusion of different
  • 17. 6 types of water consumption, by different types of dwellings, as well as leaks from water pipes. As for monthly consumption data, it is constructed by transforming quarterly data. For example, Table 2.1 shows meter readings taken between 20 April 2007 and 17 July 2008 for a particular property. There are four meter readings that were taken during the financial year 2007-2008. These readings were taken on 10 July 2007, 22 October 2007, 16 January 2008 and 17 April 2008. Table 2-1: Water rates readings for an example property for the period 2007-2008   Year    Quarter    Meter read date    Consumption   Days   Daily Consumption    2006‐07  4  20‐Apr‐2007  1255  92  13.6  2007‐08  1  10‐Jul‐2007  1355  90  15.1  2007‐08  2  22‐Oct‐2007  2205  95  23.2  2007‐08  3  16‐Jan‐2008  1511  86  17.6  2007‐08  4  17‐Apr‐2008  1391  92  15.1  2008‐09  1  17‐Jul‐2008  1252  91  13.8  Source: Based on Sydney Water data sets The recorded meter reading for Quarter 1 was taken on 10 July 2007. The previous meter reading was done on 20 April 2007, which implies that the water consumption was between 20 April 2007 and 10 July 2007. Furthermore, in some types of analyses carried out by Sydney Water, it is often more convenient to work using an alternative measure for consumption. This measure is referred to as monthly apportioned consumption. It is constructed by calculating the average daily consumption between two meter readings and rolling this up into monthly totals. Figure 2-1 illustrates this procedure. Figure 2-1 shows the average daily consumption between the meter reading dates given in the Table 2- 1. For example, between the period 20 April and 10 July 2007, the average daily consumption was 15.1 kL/day. This is calculated as the metered consumption on 10 July (1,355 kL) divided by the number of days since the previous meter read (90).
  • 18. 7 Figure 2-1: Example average daily consumption between the meter read dates Source: Based on Sydney Water data sets By multiplying the average daily consumption during a month, by the number of days in the month, the apportioned monthly consumption is obtained. From Table 2-1, it is noted that the average daily consumption during September 2007 is 23.2 kL/day. This is then multiplied by the number of days in September (30) to give an apportioned monthly consumption of 696 kL for September 2007. The calculation becomes a little more complex for the month the meter is read. For this, the monthly apportioned consumption is calculated using the average daily consumption for the two meter reads that cover that month. To illustrate take a meter read on the 22 October 2007. The average daily consumption for the period 1 to 22 October is 23.2 kL/day (calculated using the meter read taken on 22 October). The average daily consumption for 23 to 31 October is 17.6 kL/day (calculated using the next meter read taken on 16 January 2007). The average daily consumption for October is therefore calculated as 2223.2 (1-22 October) + 917.6 (23 to 31 October) = 669 kL.
  • 19. 8 The method so far discussed to calculate the average consumption of water induces a measurement error in the dependent variable and serial correlation in the error process. Hence any price elasticity studies undertaken using this processed average monthly data can lead to wrong inferences. As shown in the previous discussions, there may be several assumptions or data manipulations that were used in the construction of the data published by the water authority. External consultants who use these secondary data may not be aware of these assumptions. The lack of transparency in some of the published data or reports may make it difficult to review the assumptions and issues associated with the data. This problem is increased when the data comes from different water authority areas, each with their own assumptions and data gathering method. 2.2 Determinants of water demand Different studies use various determinants of water demand. The choice of determinants is highly dependent on the aim of the study. The key variables that influence water demand reviewed for this study are as follows:  Price variables  Weather variables  Household information  Property specific variables The above factors will be discussed in more detail in the succeeding subsections. 2.2.1 Price Grafton and Kompass (2007) and also Grafton and Ward (2007) tried to estimate real price elasticity of demand for water during the period where there was a real price change (Jun 1995 to Jun 2005). However, the price of water usage had not changed much between July 2000 and October 2005 in real terms. Warner (1996) used usage price and revenue per kL, between the years 1960 and 1994, when some level of price increase was observed. Abrams et al. (2012) conducted a study between Jun 2004 to Jun 2010 when again a real price usage increased. Barkatullah (2002) analysed the consumption
  • 20. 9 between the years 1991 and 1994 where increasing block tariffs occurred, although minimal. Thus the period between June 2005 and June 2009 represents a real price increase; hence, this would be the best period for this study. In a two tiered price structure, many studies used Nordin’s (1976) specification. The issues associated with two or more tiered tariff structures are outlined in Taylor (1975). Following Taylor’s study, Nordin introduced a difference variable referred to as the ‘rate structure premium’. It is simply computed as the difference between the total bill and the estimated bill, if the water quantity was consumed at the marginal price. Following Nordin (1976) the following price variables are the most appropriate price variables to use:  the marginal (highest) price paid for water by the household, and  the difference between the household’s total water bill and what they would have paid if all water consumption was charged at the marginal price. There are, however, a few studies, such as that by Chicoine et al. (1986), which argue that Nordin’s suggestion is unnecessary, while a few studies, such as that by Barkatullah (1996), agree with Nordin’s method. Abrams et al. (2012) used a Weighted Average Price paid per kL by each residential household and applied Nordin’s method. Warner developed two models based on ‘real average revenue per kilolitre’ and ‘real marginal price of water per kilolitre’. These two price variables depend on water usage, implying that are endogenous. However, Generalised Method of Moments (GMM) can be applied to overcome the endogenous problem. 2.2.2 Weather Variables and Seasonality Weather factors, such as temperature, rainfall and evaporation, are considered to be important factors in determining water demand. Although every water demand related study has included weather or season related variables in their models, most of them used rainfall and temperature without any transformations. The following discussions consider the few studies that analysed their weather variables and came up with better ways to include these in their models. Maidment and Miaou (1986) found that the effect of rainfall is influenced by its frequency of occurrence, magnitude and level of temperature changes. They also found that there are two distinct
  • 21. 10 blocks of temperature ranges, namely (i) 4-21 C and (ii) above 21 C, where water usage does not change much. Bamezai (1996) argues that deviation from the average value of temperature; rainfall and evaporation are a better way to represent the weather variables. Other studies use absolute temperature and absolute evaporation instead. Only a few studies had used a number of rainy days above 2mm instead of absolute rainfall data. Furthermore, Abrams et al. (2012) used weather data from thirteen weather stations observed across Sydney. In this study they have examined rainfall, temperature and evaporation as contributing to deviation from the average values. Howe et al. (1967) used summer precipitation and maximum day evapotranspiration to estimate a sprinkling demand model. Similarly, Warner (1996) used rainy days (number of rainy days from July to December), temperature (monthly average maximum December to March) and soil moisture (average bucket from October to March) to account for the weather and seasonality. However, the reasons for selecting particular months were not discussed in the paper. Most of the existing studies reviewed use dummy variables to represent seasonality and almost all of them found that summer price elasticity of demand is negatively higher than the winter price elasticity. Abrams et al. (2012) handled the seasonality by including dummies to handle the four seasons. There are however studies such as those by Warner (1996) that did not bother including seasonality as the data used was yearly where all seasons are included. In summary, seasonal dummies can be specified by seasons, namely: winter, spring, summer and autumn using a quarterly dummy. Moreover, seasonality can be captured by specifying a monthly dummy. Alternatively, since the meter readings are not neatly aligned with a season or month; it is also possible to create variables to indicate how much a season (or month) is covered by a particular meter reading. 2.2.3 Household information The household size is an important factor in determining water consumption. However, many studies did not include this important variable due to lack of available data. Obtaining household income is even harder, as people tend to be wary of discussing their income. The study by Abrams et al. (2012) specified a ‘difference’ model where the household size is assumed to be time-invariant. Conditions placed on the population to satisfy this assumption could introduce some selection bias and survival bias. To address this, Martinz-Espineira (2003a) used a variable to
  • 22. 11 indicate the proportion of the population over sixty-four years and under nineteen years. Warner (1996) did not include household size as a determinant as his dependent variable was per capita consumption of water. Nauges and Thomas (2000) pointed out that household composition also significantly influences the water consumption. For example they found that households with young children or retired persons tend to use a lot more water than other households. In addition, Abrams et al. (2012) calculated the mortgage stress using household income and mortgage from aggregated ABS data. However, using the aggregated data at micro level of analysis may introduce some measurement errors. In 1996, Warner used ‘real average weekly male earnings’ as a factor however in 2011 male earnings alone may not represent the actual household income. Worthington and Hoffman (2008) concluded that the estimates of income elasticity in the literature are almost universally income inelastic, while noting sample or specification bias problems. They suggest that a more complex model allowing for longer run transitions should be developed. Joanne et al. (2012) states that single occupant households with a retired resident used 70% more water than a working age occupant as the retired resident spends more time at home. 2.2.4 Property specific information Property specific information, such as property size, pool information and water efficiency participation also play an important role in determining the level of water consumption. Abrams et al. (2012) used the lot size as a proxy for both the property and garden size variables. However, by using the ‘first differences’ technique, this time-invariant factor does not have any effect on the model estimates. They instead used the lot size to cluster their housholds into different groups. In the earlier study by Warner (1996) these variables were not specified. Again, omitting these property related variables may not cause any serious problems if the characteristics of properties do not change over the study period. Abrams et al. (2012) removed any properties which constructed pools during the study period. This process may result to selection bias as the exclusion of households with other odd factors affecting water use can affect the modelling results. Others argue that this is captured by the error term in the model. Worthington and Hoffman (2008) suggest that income can be used to proxy other normal and
  • 23. 12 luxury goods (e.g., swimming pools and in-ground garden irrigation systems.) associated with household water consumption. Information, such as the existence of pools and participation in water efficiency, can be obtained from Sydney Water. Worthington and Hoffman (2008) suggested that swimming pools may indicate that these households are relatively wealthier than other households. It can also indicate a higher chance of children living in these households. Similarly, participation in different water efficiency programmes may indicate different household characteristics. For example, households which participated in the ‘Love Your Garden’ programme may be keen gardeners. Based on these arguments it is therefore useful to segment the properties based on such information and develop separate models. 2.3 Functional forms The nature of the function to estimate water consumption is not well defined in the literature. There are number of functional forms that are used to specify water demand or calculate water demand elasticity. It is however evident that it is not possible to develop economic models independent of functional form assumptions, as the parameter estimates may well be very sensitive to these assumptions. Zarembka (1968) has pointed out that economic theory provides little guidance on appropriate functional forms for demand functions. As there is no consensus on the type of functional form to use, it is left to the researchers to choose the relevant functional form that answers their specific problem. Many studies reviewed such as Nauges and Thomas (2003) assume a certain functional form without explaining the reasons behind their choice. Some studies such as those of Agthe and Billings (1980) selected models based on which one gave the best statistical results while Howe and Linaweaver (1967) defend their choice of double-logarithmic functional form by arguing that any of the theoretical considerations fail to specify a unique functional form for instance those by Williams (1985) and Dandy et al. (1997), as it yields direct estimates of elasticity. Like the linear function, the double-log function also leads to a constant-elasticity of water demand. However, as pointed out by Al-Quanibet and Johnston (1985), this functional form lacks consistency with utility theory. Furthermore, double- log function is curvilinear which means that there is no such direct concept as a choke price.
  • 24. 13 As water is an essential commodity, the consumer demand theory indicates that a minimum amount of water demand is expected even at a very high price. Similarly, consumers do not use an infinite amount of water even if the price of water is zero. Al-Quanibet and Johnston (1985) and Gaudin et al. (2001) both used the Stone-Geary specification that is able to address this limitation but the functional form can be complex and difficult to interpret. A semi-log model on the other hand is consistent with economic theory that indicates that consumers are more sensitive to change in price when the price is high. Moreover, the semi-log function is curvilinear which means that there is no such a direct concept of a choke price. Semi-log models have also been used in many studies, such as Arbues et al. (2003) and Barberan and Villanua (2000). Further, Abrams et al. (2012) argue strongly that the semi-log functional form is the specification for residential water demand. 2.4 Summary As discussed in this chapter, the nature of the data set used by the analyst will greatly influence the results. Most Sydney based studies reviewed used ‘bulk’ water demand but this may result in aggregation bias due to process of combining different types of water consumption, different types of dwellings and inclusion of the leaks from water pipes. It is therefore suggested more appropriate to use the metered water consumption data instead of the “bulk data”. Furthermore, it is evident that there is a difference between the processed ‘monthly’ water demand data and quarterly recorded metered data. Very few studies reviewed have chosen their period of analysis where the change in price was minimal. It is important to select the period where the price was stable prior to the period of analysis in order to estimate the impact of price change. Under more than one-tier pricing structure, the price variable becomes endogenous. Many studies used Nordin’s (1976) specification to handle this endogeneity issue although a few studies also argue that Nordin’s suggestion is unnecessary. Although most of the water-demand related studies have used rainfall and temperature as an explanatory variables oftentimes these variables are not transformed. Few studies such as Bamezai, A. (1996) used deviation from the average values of weather variables. Seasonality was mainly captured by dummy variables.
  • 25. 14 The household size is arguably the most important factor in modelling water demand. However, it is not easy to obtain information on the household size or household income. A model with a ‘difference’ specification can solve this misspecification problem as long as these variables are time-invariant. Noticeably, property specific information such as property size, pool information and water efficiency participation is omitted in most studies. The study undertaken by Sydney Water, Abrams et al. (2012), included these factors as they had access to these data held by Sydney Water. However, the said study did not develop separate models for different consumer segments. Unfortunately, the type of function to estimate water consumption is not well established in the literature. The double-log model specification is often used in elasticity studies as it yields direct estimates of elasticity. The second most popular specification is the semi-log model. The semi log model indicates that consumers are more sensitive to change in price when the price is high. It is also curvilinear; hence, there is no direct concept of a choke price.
  • 26. 15 C h a p t e r 3 - T H E O R E T I C A L F R A M E W O R K 3.0 Introduction This chapter discusses the development of the model applied in this study. Section 3.1 presents the general form of the model. The resulting model specified for each segment can vary slightly, based on theoretical and practical considerations as well as the results of the various diagnostic tests. Section 3.2 looks at commonly included variables in literature and evaluates the suitability of each variable transformation performed to correctly specify the model. Further, it discusses the reasons as to why the factors are selected and ways to minimise the specification errors. Specification error may occur due to errors of omission and errors of inclusion. Errors of omission happen when crucial variables are omitted from the model. On the other hand, errors of inclusion occur if useless variables are included in the model. Section 3.3 investigates commonly used functional forms and argues why a semi-log model specification is used in this study to calculate the demand for water. There are three commonly used functional forms in literature, namely: linear, double-log and semi-log. Each of these functional form represents different outcomes for a given price change. Hence, it is important to use the most appropriate functional form in specifying demand. Model formulation requires the specification of a functional form that best defines the relationship between the explanatory variables and the predicted variable. As Zarembka (1968) noted, there is no economic theory that provides guidance on choosing the appropriate functional for demand functions. As shown above, the functional form of water consumption and water price determine the basic shape of the demand curve for water.
  • 27. 16 3.1 Demand function A general dynamic model specification for this study is expressed as, )1.3( springwinterautumnsummer nsrestrictiorainfallnevaporatio pricepricepricepricell 4321 765 34231214 itiit itttit ttt ititititititit u u nCnC          where, Cit = Quarterly water consumption of household ‘i’ at time period ‘t’ Cit-4 = Quarterly water consumption of household ‘i’ at time period ‘t-4’ priceit = Weighted average water usage price of household ‘i’ at time period ‘t’ evaporationit = Deviation from the average observation of evaporation at period ‘t’ rainfallit = Deviation from the average observation of rainfall of household ‘i’ at time period ‘t’ restrictionit = Variable to indicate how many water restriction period is covered of household ‘i’ at time period ‘t’ summerit = index to indicate length of summer period covered of household ‘i’ at period ‘t’ autumnit = index to indicate length of autumn period covered of household ‘i’ at period ‘t’ winterit = index to indicate length of winter period covered of household ‘i’ at period ‘t’ springit = index to indicate length of spring period covered of household ‘i’ at period itu = Error term ηi = Time-invariant household specific effects εit = Random noise It should be noted that the model may vary for each consumer segment. As explained in Section 3.3, water consumption (and lags of water consumption) is transformed into natural logarithms. The auto- regressive, distributed lag (ARDL) regression model is specified to determine demand for potable water. The dynamic nature of the regression model allows for lagged responses to changes in price and can be used to estimate the time it will take consumers to fully adjust to changes in prices.
  • 28. 17 3.2 Water demand and explanatory variables It is impossible to capture all factors affecting household water consumption. For this study, the following variables that are identified in the literature to be significant variables in determining demand for potable water are included in the model specification.  Historical water consumption patterns  Price of water  Seasonality and weather  Property size and garden size  Household size and disposable income  Government rules and regulations  Adaption of water efficiency practices  Cultural factors 3.2.1 Water demand defined Generally, water usage data can be obtained in two forms – bulk water usage and metered water usage. Bulk water is measured as the total amount of water supplied by a water authority through the water distribution network. In Sydney, it is mainly the Sydney Catchment Authority (SCA) that supplies water to Sydney Water. The amount of water measured at SCA’s flow meter is referred to as ‘Bulk water’ in this study. Metered water is the amount of water used by each property and is typically measured over a ninety day-intervals in Sydney. Although many of the previous studies reviewed have used bulk water data in their analysis, it is unsuitable for modelling the impact of water price for places like Sydney, for the following reasons: Firstly, bulk water is the aggregated water use of all users. It is not possible to separate different user groups such as government organisations, commercial properties, industries, houses and units. Water usage is very different across these different user groups and the fundamental reasons for their water consumption varies hence a single model simply cannot capture the nature of demand of each of these sectors.
  • 29. 18 Figure 3-1: Water usage spilt in 2009-10 between different type of users Source: Derived from Sydney Water data sets As shown in Figure 3-1, only about 75% of potable water is consumed by residential dwellings. The remaining 25% of potable water is consumed by various industries, which have a completely different consumption pattern to residential potable water demand. Secondly, there has been a systematic reduction in industrial water usage and leaks. Sydney Water continuously invests in fixing leaks, like many other water authorities. Figure 3-2, for example, shows that there is a steady reduction in leakage and industrial usage in Sydney over a ten-year period. Any reduction in bulk water demand due to leak reduction programs need to be considered if bulk water is used. Otherwise, these water usage reductions would be allocated to other variables, such as price. Government & Other 8% Commercial 10% Industrial 9% Residential Single dwelling 51% Residential Multi-unit 22%
  • 30. 19 Figure 3-2: Industrial usage and leakage over time Source: Based on Sydney Water data sets Thirdly, it may not fully account for the impact of water efficiency programs introduced during the study period. Sydney Water has been implementing water efficiency programs for over a decade. Figure 3-2 shows the water savings achieved through water efficiency programs. These programs are estimated to reduce the water demand by around 90 GL per year. The impact of the water efficiency programs on bulk water cannot be ignored. Figure 3-3: Impact of water efficiency programmes Source: Sydney Water – Water Conservation and Recycling Implementation report 2010-2011 0 20 40 60 80 100 120 140 2001-02 2002-03 2003-04 2004-05 2005-06 2006-07 2007-08 2008-09 2009-10 GL/year Leakage Industrial 0 20,000 40,000 60,000 80,000 100,000 120,000 140,000 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Year Ending June Watersavings(ML/yr) Regulatory measures Recycled water Leakage reduction Business program Residential outdoor Residential indoor
  • 31. 20 Finally, as discussed earlier, artificially created ‘monthly consumption’ is not suitable for this study either. Due to all the shortcomings listed here, this study uses actual recorded water consumption for each property. However, care needs to be taken to account for the differences in timing of individual meter reads, as residential water meters are read four times a year on a rolling basis. 3.2.2 Price of water Sydney Water bills consist of water usage price, the service price and also the total water bill. These prices are set by The Independent Pricing and Regulatory Tribunal (IPART). Sydney residential properties are charged under two categories: fixed charge and usage charge. Fix charge comprises of ‘water service’, ‘wastewater service’ and ‘stormwater service’ charges. Table 3.1 shows the pricing structure for the financial year 2010-2011. Table 3-1: Sydney water price structure 2010-2011 Charge Type  Amount  Water service (Fixed)  $31.30  Wastewater (Fixed)  $129.29  Stormwater service (Fixed)  $11.92  Usage per kL (Variable)  $2.01  Source: Sydney Water Financial report June 2011 It can be argued that households react to changes in the total amount they pay when they receive their water bills. In Sydney, water bills are generally dominated by usage charges rather than service charges, as shown in Figure 3-4.
  • 32. 21 Figure 3-4: Service charge and usage charge distribution for a household uses 240kL per year (in real terms) Source: Constructed based on Sydney Water data sets It is also important to note that the service charge has not increased so much, compared to the usage charge, as shown in in Figure 3-5. Thus it makes more sense to use water usage prices as an explanatory variable rather than service prices or total water bills. Figure 3-5: Inflation adjusted real price of fixed charges and usage charges over time Source: Constructed based on Sydney Water data sets Figure 3-5 further shows that although there is no change in water usage price (in real terms) between June 1995 and June 2005 it started to increase after June 2005. In addition, in this study, a second tier 0 100 200 300 400 500 600 700 800 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 Cost Usage Tier 1 in c/kl Usage Tier 2 in c/kl Fixed Charges in $
  • 33. 22 price for residential consumption is applied if the water consumption level is above 400 per kL per household per year. Sydney was under a two-tier tariff water price structure from 1 October 2005 to 30 June 2010. Unfortunately, in a two-tier pricing system, the price paid is no longer exogenous, as it depends on the level of consumption. As price is the key independent variable for this study, it is important to specify this correctly. In this study, the procedure applied by Abrams et al. (2012) is adapted. In that study, they used the GMM estimation technique with appropriate instruments to address the endogeneity problem. The endogeneity problem arises when the dependent variable (in this case water demand) has a causal effect on one or more of the explanatory variables (for example price). Specifically, a GMM is specified using a Weighted Average Price (total usage charges divided by total water use is used in the model, together with Tier1 price and Tier2 price serving as instruments. Tier1 and Tier2 prices are an appropriate choice of instruments, as these two are highly correlated to the Weighted Average Price but are exogenous to water consumption. 3.2.3 Seasonality The effect of changes in season is different from one household to another. For example, a property with large gardens may use more water than a similar property with smaller garden during spring. Properties with pools may use more water in summer than similar properties without a pool. Larger households may use more water compared to smaller households during winter due to differences in the number of showers and washings. Almost all existing studies reviewed used dummy variables to account for seasonality. However, as water consumption is only measured quarterly, seasons do not neatly align with meter reading. Very often a single meter reading period can include two different seasons. Hence for this study monthly variables were specified to capture monthly indices. For example, if a property is read on 10 July 2007 and 22 October 2007, then monthly indices were calculated as follows. Value of variable ‘Jul’ = (31-10)/31=0.68 Value of variable ‘Aug’ = (31-0)/31=1.0 Value of variable ‘Sep’ = (30-0)/30=1 Value of variable ‘Oct’ = (22)/30=0.73
  • 34. 23 This particular observation includes 68% of July, 100% of August, 100% of September and 73% of October. Similarly, it is also possible to calculate four seasonal variables, namely summer, autumn, winter and spring, using the above monthly indices. The seasonal variables are calculated as follows. Value of variable ‘Summer’ = (variable ‘Dec’ + variable ‘Jan’ + variable ‘Feb’)/3 Value of variable ‘Autumn’ = (variable ‘Mar’ + variable ‘Apr’ + variable ‘May’)/3 Value of variable ‘Winter’ = (variable ‘Jun’ + variable ‘Jul’ + variable ‘Aug’)/3 Value of variable ‘Spring’ = (variable ‘Sep’ + variable ‘Oct’ + variable ‘Nov’)/3 Interpretation of seasonal variables is similar to that of the monthly variables. Each set of seasonal variables were experimented during the modelling exercise. However, the choice of the set of seasonal variables is based on the modelling diagnostics for each segment. 3.2.4 Weather conditions In most of the literature reviewed, temperature, rainfall and evaporation are considered important factors in determining water consumption. Weather changes cause short-lived fluctuations in demand rather than underlying changes. Discretionary water use, such as watering of gardens or washing of cars, tends to be more price responsive but at the same time influenced by the weather conditions. Water demand in summer is expected to increase with temperature and evaporation but decrease with rainfall. In Sydney, weather patterns vary across regions. For example, the eastern suburbs receive high rainfall and record low temperatures compared to the western suburbs. Hence it is important to obtain weather observations from multiple weather stations spread across Sydney. The literature further notes that, there are many ways these weather variables can be specified in the model. The following variations of weather variables were calculated and experimented during the modelling exercise.  Average weather observations for each recorded meter period for each property  Based on the paper by Bamezai (1996), deviation from the average value of weather variables  Based on Maidment and Miaou’s (1986) study distinct blocks of temperature ranges, namely (i) 4-21 C (ii) 21-35 C and (iii) above 35 C and also number of rainy days with more than 2mm rain were also taken into account.
  • 35. 24 The final weather variable specification for each segment was highly dependent on the diagnostic test results. The Bureau of Meteorology (BoM) holds most weather data such as rainfall, temperature and evaporation. Unfortunately, this data set is far from perfect, as there are many missing observations. Figure 3-6: Map of Sydney weather stations used for this study Source: Google Map Figure 3-6 shows the location of weather stations at Prospect (A), Richmond(C), Terry Hills (D), Sydney Airport (E) and Holsworthy (F). Weather stations at Nepean and Warragamba are not shown on the map, to make the map readable. Weighted average weather values are calculated for each property, based on these eight stations, using distance as weight.
  • 36. 25 3.2.5 Government rules and regulations Sydney water internal studies indicate that government rules and regulations regarding water use can influence household water use especially outdoor water usage. The study is conducted during the period when Level 2 and Level 3 drought restrictions were in place. Hence dummy variables were specified to account for these restrictions to separate these two periods. The season variables are specified (monthly variables), as discussed in Section 3.2.3. Likewise water restriction variables are also calculated and specified as earlier discussed in Section 3.2.3. 3.2.6 Property size and garden size The size of a property and garden also play an important role in determining the level of water consumption. Sydney Water’s internal data analysis shows that larger properties tend to have more bathrooms hence less pressure to have shorter showers. In addition, there is a high chance of social functions happening or guests staying over in larger properties. Furthermore, households with larger the garden tend to have greater outdoor water consumption. In this study, to specification of these factors are not included as they are time invariant and not important in a differenced or fixed effect model. However, these factors are used to classify consumer segments. 3.2.7 Household information The Census of Population and Housing conducted by the Australian Bureau of Statistics (ABS) is the main source of small area socio-demographic data in Australia. It is conducted every five years. The Census is a count of the population and dwellings in Australia, with details of age, gender and a variety of other characteristics (ABS, 1996). The Census Collection District (CCD) is the smallest geographic area defined in the Census. There are around 225 dwellings in each Sydney CCD and there are more than 2000 CCDs defined in the Sydney Water service area. The latest household information of each CCD is as of August 2006. The household size is arguably the most important factor in terms of determining the level of water consumption. It is relatively safe to assume that the household size does not change in a short period of time. Generally, people change their residence if their household composition changes. For example,
  • 37. 26 couples without kids tend to move to a bigger house when they have children. Older parents tend to move out of their large homes when their children leave home. Studies show that households with more disposable income tend to be less responsive to increase in the price of water, as it requires a small portion of their disposable income. They may also have newer or more efficient water appliances and fittings. On the other hand households under financial stress may not have any discretionary water use; hence the price change may not have any effect. They however may not be able to take part in water efficiency activities, such as installing rain water tanks, as they cannot afford them. The Worthington and Hoffman’s (2008) study indicates that the estimated income elasticity of demand for water is inelastic. Applying any averaged household information, such as household size and household income, to individual households, introduces measurement error. For this study, households were carefully filtered to make sure that their household size did not change. Given the difficulty in accurately measuring household size and household income related data, this study used these factors to segment the households. Also, these time invariants do not have any effect on model estimates if the ‘first difference’ model or fixed model is specified. 3.2.8 Adaptation of water efficient appliances or activities Adapting water efficiency activities can reduce households’ demand for water. This could be achieved by installing water efficient appliances or substitutes, such as rainwater tanks. Abrams et al. (2012) included water efficiency programs in their models and estimated the impacts of those water efficiency programs. Hence in this study instead of specifying variables in the model, the information about water efficiency programs are used to create a consumer segment. This study has a separate segment for households which participated in any water efficiency programs outside the analysis. 3.2.9 Habits and culture “We are, all of us, creatures of habit, and when the seeming necessity for schooling ourselves in new ways ceases to exist, we fall naturally and easily into the manner and customs which long usage has implanted ineradicably within us.” - Edgar Rice Burroughs, The Beasts of Tarzan. A person used to
  • 38. 27 having a twenty-minute shower is likely to continue their behaviour regardless of change in water price, if it is small. Furthermore, the literature reveals that as a result of the force of habit, individual households may only change their consumption after sometime. For example, Level 3 water restrictions were in place for about three years. Due to this level of restriction, the average household water usage dropped dramatically. However, they did not bounce back once the restrictions were lifted and this may be partly due to the fact that households have permanently changed their habits. Future expectations may also change a household’s current behaviour, for example weather forecasts may play a factor in deciding whether to water a garden on that day or not. This may cause a weak exogeneity problem in running the model. Dynamic regression models can handle this weak exogenous regressors (lagged dependent variable) problem. Abrams et al. (2012) address habit formation through the dynamic specification of the econometric model, by including the previous period’s consumption. This study also follows the same procedure, as it has been proved to be valid. However, the number of lagged consumption variables is based on the results of model diagnostics of each consumer segment. 3.2.10 Other water uses Major water use can include many different water-related activities, such as filling up a swimming pool or using a water sprinkler. Excluding information on the presence of pools or the use of water sprinklers can contribute to the misspecification of the model. Swimming pool is not a variable specified in the model but rather used to create a separate consumer segment based on the existence of swimming pool information.
  • 39. 28 3.3 Functional form of demand The linear water demand functions are the easiest to use, as no transformation is required. The linear functional form implies that the slope which measures the change in quantity demanded due to a change in price level. Mathematically, the general form of the model can be specified as follows:   0,  iiiiii EXY  (3.2) where, iY = Dependent variable water demand by household i iX = Independent variables Elasticity of Y with respect to X is                            )( )( )( )( yE xE y x x y xE yE i (3.3) Based on the above equations, β measures the change in y per unit change in x. Hence, it also implies that the lower the price, the less sensitive consumers are to changes in prices. In reality, the linear shape cannot be assumed to apply to the whole range of the water demand curve. Otherwise, a choke price can be found at which consumers do not consume any water. Clearly that is not the case, as water is a necessity commodity. Linear form may be applicable in calculating ‘point elasticity’ where only a section of price change is considered. Many studies reviewed used the double-log model specification as it yields direct estimates of elasticity. Unlike the linear function, the double-log function leads to a constant price-elasticity of water demand although the double-log function does not have the choke price problem as it is curvilinear. Unfortunately, it is not consistent with the utility theory. Hence, the semi-log model specification is used. The general form of the model   0,)( 0  iiiii EXYLn  (3.4) The above model is consistent with economic theory, which indicates that consumers are more sensitive to a change in price when the price is high. It is also curvilinear, which implies that there is no
  • 40. 29 such direct concept of a choke price. Given the difficulties associated with measuring income at the household level, the semi-log model specification appears to be an attractive starting point over the Stone-Geary form and the double-log model. In summary, a semi-log dynamic model is specified and the GMM estimation procedure is carried out for this study. The dependent variable water demand is therefore transformed into natural log form. The weighted average usage price is calculated and included as the key price variable. Due to exogenous problems, the Tier1 price and Tier2 price are used as instruments for the GMM estimation technique. Commonly used variables, which are difficult to measure at household level or are time invariant were used to create the consumer segments. Dummy or proxy variables were also used to take into account the weather, season, water restrictions and consumer habits.
  • 41. 30 C h a p t e r 4 - M E T H O D O L O G Y 4.0 Introduction The principal method of analysis applies techniques in econometrics that use complex panel data on various segments of the residential market. Panel data is a combination of time series and cross- sectional data. It makes use of individual water consumption and relevant explanatory variables of a sample of properties over time. The key advantage of using panel data is the ability to obtain more accurate estimates of the impact of price and other factors as compared to those obtained from cross- sectional or time series data only. Furthermore, the technique makes it possible to control for the effect of time-invariant unobserved variables. Water consumption patterns are expected to be different, depending on dwelling type, tenancy, participation in Sydney Water’s water efficiency programmes, family financial status and property size. Fourteen residential consumers segments were created and Section 4.1 covers these segments in detail. Section 4.2 defines the study area and is referred to as the Sydney Water service area. This section also discussed the reason behind the selection of the study period. Finally, Section 4.3 discusses commonly used estimation techniques, such as ordinary least squares (OLS), maximum likelihood (ML) and GMM and argues why GMM is chosen for this study. 4.1 Residential consumer segments As described in Chapter 1, the various water consumer segments of Sydney households are formed based on a combination of dwelling, economical, financial, geographical and social characteristics. Segmenting the residential water consumers may help reduce the heterogeneity among households. It is important for a utility company to strengthen their bond with consumers today to create a better business for tomorrow. The reasons as to why any business may need to segment their consumers are as follows:  Consumer needs can be better matched  Increased return on investments  Increased opportunities for future growth  Customised and targeted communications
  • 42. 31 As for Sydney Water, this is a starting point to see how different consumer segments are reacting to the change in water price. The following factors are identified to be important aspects in segmenting consumers based on the interview of various Sydney Water managers:  Type of dwelling  Economic status  Financial status  Geographical location  Social status More specifically, this study will identify various consumer segments based on the following combinations of household characteristics:  Property structure  Property ownership  Water efficiency participation  Financial hardship  Household income  Lot size  Surname origin  Pool ownership Figure 4-1 shows how the above ‘factors’ were categorised. It further reveals that there are potentially 72,000 consumer segments that can be created by utilising all these categories. However, only fourteen consumer segments were considered for the analysis as they were identified as being important to the business and they also had enough samples for modelling purposes.
  • 43. 32 Figure 4-1: Socio-Economic-Demographic Factors From Figure 4-1, the base category is identified to have the following characteristics:  House (single detached dwelling)  Owner occupied and the owner has only one property  Not participated in any water efficiency program  Did not receive any financial assistance  Located in average income census collection district  Has average lot size  Owner has a surname which is not popular East Asian or Islamic or Indian sub-continent name  Does not have a swimming pool All other segments were similarly created by changing only one category from the base segment. Thus, each segment is mutually exclusive as shown in Appendix 5 (Consumer Segments).
  • 44. 33 4.2 Study area and Study period Sydney Water covers more than 12,000 square kilometres of area, servicing more than four million residences. Figure 4-2 shows the Sydney Water area of operation, which goes beyond Sydney to areas such as the Blue Mountains and Wollongong. Figure 4-2: Sydney Water area of operations Source: Sydney Water Website
  • 45. 34 The Sydney Water billing database has approximately 1.7 million residential dwellings and 167 thousand non-residential properties. There are almost fifty different classifications of properties. Even within a property classification, such as single dwelling (detached houses), properties are highly heterogeneous. The main reasons for this heterogeneity are location, income and attitude towards water use. Sydney Water supplies three types of water: filtered, unfiltered and recycled. In 2009-10, it supplied residential properties with around 331 gig litres (GL) of potable water, and 2 GL of recycled water. Figure 4-3 shows potable water demand (bulk demand) per day in mega litres, since 1999. Coloured background (Vol, L1, L2 & L3) indicates the level of water restrictions were in place. The figure clearly shows a reduction in water consumption between July 2005 and June 2009. This reduction could be due to a number of reasons including heavy water restrictions, effectiveness of water conservation programmes and water price increases. Figure 4-3: Timeline (bulk demand, real price increases, water conservations and levels of water) Voluntary restrictions were introduced in early 2003 when the residents were asked to reduce their water consumption. On the 1st of October 2003, Level 1 water restrictions were put in place. During this period, residents were prohibited from hosing hard surfaces, and using sprinklers or water systems. More restrictions (Level 2) such as limited garden watering days (Sundays, Wednesdays and Fridays) were further introduced on the 1st of June 2004. Finally, from the 1st of June 2005 until the 21st of June 2009, the strictest Level 3 restrictions were imposed allowing for only two days of watering the gardens.
  • 46. 35 Despite of all these restrictions, the Sydney’s real water usage price (adjusted to inflation) did not change much for a long period (July 1999 to September 2005) as illustrated in Figure 4-3 (the red line). After this stable price period, there were only three real price increases, which occurred in October 2005, July 2008 and Jul 2009. When the Level 3 water restrictions were removed in July 2009, the water price also increased which would make it difficult to estimate the water reduction for one factor in isolation of the other factor. Hence the study period for this project is from June 2004 (start of the Level 2 restrictions) to June 2009 to capture the various restrictions and changes in water price. 4.3 Estimation methods The Ordinary Least Square (OLS) method is a popular method for estimating the unknown parameters in a linear regression model. OLS minimises the sum of squared vertical distances between the observed responses in the dataset and the responses predicted by the linear approximation. The OLS estimator can be applied only under the following conditions:  the regressors are exogenous  there is no multicollinearity  the errors are homoscedastic and serially uncorrelated OLS can provide minimum-variance mean-unbiased estimation if the above conditions are met. However, the model specified for this study has some lag values of regressors and has unobserved heterogeneity that is inherent in the data; hence, OLS-type procedures would produce biased estimates. For example, consider the following one way error component model. 1 … 1 … . . 4. 1 where is usual individual effects and is usual error term. Unfortunately in the above equation, is correlated with both and which makes OLS biased and inconsistent.
  • 47. 36 To address these issues, the study explored panel data analysis method. The panel data analysis has two major specifications; the fixed effect and the random effect models. Fixed effects models or within-transformation is one of the popular ways to sweep out the individual effect . However, Nickell (1981) proves that fixed effects estimators are biased and inconsistent for panel data sets with large T (number of time periods). Further, Robertson and Symons (1992) argue that even when T is large, fixed effects estimates of dynamic panels with heterogeneous coefficients on the lagged dependent variable or auto-correlated regressors are inconsistent. Random effects models assume that the unobserved variables are uncorrelated with all the observed variables. Random effects estimators are more efficient than fixed effects estimators, provided the assumptions underlying are satisfied. In order to satisfy the assumptions, it is necessary that the individual specific effects be orthogonal to the other covariates of the model. The Hausman specification test is used to test these assumptions. This is done by first running the random effects estimator, then the fixed effects estimator and finally conducts the Hausman test. For this study, it is not wise to assume that the unobserved variables are uncorrelated with all the observed variables. Another popular demand estimation specification is the Maximum Likelihood (ML) estimator. OLS is the ML estimator, if the additional assumption that the errors be normally distributed is met. The ML estimator selects values of the model parameters that maximise the likelihood function that gives the observed data the greatest probability. However, ML is not employed in this study, as it requires a complete specification of the model and its probability distribution. More recently, another estimator called the Generalized Method of Moments (GMM) has become popular. GMM was developed by Lars Peter Hansen in 1982. It is more effective when the ML estimation cannot be used, that is, if the full shape of the distribution function of the data is not known. The GMM estimators are consistent, asymptotically normal, and efficient. However, the GMM estimator requires the specification of a set of moment conditions which the model needs to satisfy. These moment conditions are functions of the model parameters and the data, where their expectation is zero at the true values of the parameters. The notion behind GMM is to estimate the parameters by matching population moments with the appropriate sample moments. This method simply minimises a certain norm of the sample averages of the moment conditions. GMM is frequently used in dynamic panel data modelling studies in different parts of the world. Nauges and Thomas (2003) used GMM estimation on French council water data. Similarly, Musolesi
  • 48. 37 and Nosvelli (2007) used GMM on Italian municipality water data. On the other hand, Martinez- Espineira (2007) used the dynamic approach for a study in Seville in Spain. After a thorough consideration of the strengths and weaknesses of each of the modelling techniques, GMM estimation is deemed the most appropriate for this study. In addition GMM estimation will capture the two-tier water usage price applied in Sydney since October 2005 in this study. Tier1 and Tier2 prices are used as instruments for Weighted Average Price in order to overcome the endogenous problem. In summary, there were fourteen segments of Sydney Water residential consumers classified based on social, economic and geographical characteristics. The study area was restricted to the Sydney Water service area and the study period was only for the period July 2004 to July 2009, as this period experienced water usage price increases. The GMM estimation technique was employed to overcome the endogenous problem when using the price variable and the lagged dependent variable as estimators.
  • 49. 38 C h a p t e r 5 – D A T A D E S C R I P T I O N 5.0 Introduction The data used for this study mostly comes from Sydney Water, which supplies water, wastewater, recycled water and some stormwater services to over 1.5 million residential households and over 100,000 non-residential properties. Sydney Water also has information about each property location, property size, dwelling status (whether it is owner-occupied or tenanted) and participation in water efficiency programs. Other information, such as household size or household income, which is not available at a household level at Sydney Water, was obtained from the ABS at CCD level. The aim of this study does not require selecting representative properties as the primary purpose is not to forecast the water demand but to calculate the price elasticity of demand using the current data. Section 5.1 briefly shows where relevant data sets were sourced from while Section 5.2 explains what restrictions were placed to obtain the most suitable properties for this study. Table 5-1 shows the overall statistics of the dataset complied to perform the panel analysis. 5.1 Data sources Various data sets were collected from Sydney Water (SW), the Australian Bureau of Statistics (ABS), the Bureau of Methodology (BoM) and the Sydney Catchment Authority (SCA). The diagram below shows datasets at top level, their sources and the information that they contain. Figure 5-1: Data Sources
  • 50. 39 As Figure 5.1 shows, the compilation of various information is a complex process that requires diligence and through knowledge of each data sets. The data set ‘Property’ is a product of combining over one hundred different tables that sit across Sydney Water databases. This data set is used to obtain information about each property, such as their property type (house, townhouse, duplex, units or flats), their geographical location, their tenancy status (rented or owner-occupied), water efficiency programme participation and their sales history. Due to data security reasons the detailed information about these data sources are not published. Another point to note is that data management is an ever evolving process. Hence even higher level information may not have any value after a year or two. The data set ‘Economy’ is compiled using Sydney Water data tables and ABS data tables. This contains information about water prices and other economy-related data at CCD level. Past pricing information is obtained from Sydney Water with all other variables were obtained from ABS. The data set called ‘Sydney Water meter reads’ contains the actual meter reading for each property in row format. Detailed knowledge about how the billing system works is essential to correctly use this database. The data set labelled ‘Weather’ contains temperature, rainfall and evaporation obtained from seven different weather stations across Sydney as presented earlier. Figure 3-6 shows the location of these weather stations on a Google map. The Bureau of Meteorology (BOM) & the Sydney Catchment Authority (SCA) compiles these data sets.
  • 51. 40 5.2 Selection of Properties It is best to have a ‘strongly balanced’ data set for a panel study. Hence, the following restrictions were placed to ensure that all properties have data for all periods.  Properties should have valid water consumption throughout the study period.  Property type (such as house or townhouse) should not have changed during the study period.  Properties should have been read 27 times within the study period. The following restrictions were also placed to satisfy the assumption that the household size did not change during the study period.  Properties should not have been sold during the study period.  Water consumption should not have increased by double or reduced by half from one recorded meter period to the other.  Very large properties (> 1000 square meter) or very small properties (<100 square meter) were removed.  Properties should have more than 0.5kL per month water consumption at any point in time. After applying above mentioned filters, the following restrictions were applied to maintain data integrity.  Properties which received any water other than potable water, such as recycled water, were removed.  Properties with water meters servicing more than one dwelling were removed.  Properties without any location information were removed.  Any properties that were read within 70 days or more than 110 days were removed. At the end, properties which fall under the consumer segments that this study is interested in were selected for the panel analysis. Basic descriptive statistics were obtained using the SPSS statistical package. The key variables and their descriptive statistics are listed in Table 5.1.
  • 52. 41 Table 5-1: Descriptive statistics for all segments Variable Name Description Min. Max. Mean Std. Dev. Unit Subject A subject identifier unrelated to the property 1 28473 14237 8219.59 property Period An identifier for each meter reading period undertaken 1 22 14 7.79 quarter Property area Property land size 100.71 1000 576.30 135 sqm Consumption Daily potable water consumption kL .01 11.24 0.65 0.38 kL/day Weighted Avg Price The average price paid by the houses during the meter reading period 114.23 200.19 148.58 26.44 Cent/kL Restrictions L2 Proportion of days spent in Level 2 restrictions during the meter reading period 0.00 1.00 0.20 0.39 % Rain The average daily rainfall experienced by the houses during the meter reading period .31 5.31 2.12 0.96 mm Temperature The average daily temperature experienced by the houses during the meter reading period 17.02 29.94 23.54 3.65 °C Evaporation The average daily evaporation experienced by the houses during the meter reading period 1.50 6.22 3.72 1.31 mm Rain deviation The deviation from the daily average rainfall experienced by the houses during the meter reading period -1.83 3.14 -0.01 0.87 mm Temperature deviation The deviation from the daily average temperature experienced by the houses during the meter reading period -2.26 1.71 0.00 0.73 °C Evaporation deviation The deviation from the daily average evaporation experienced by the houses during the meter reading period -1.17 .74 0.03 0.31 mm Jan The proportion of the meter reading period in January 0.00 .43 0.08 0.12 % Feb The proportion of the meter reading period in February 0.00 .43 0.09 0.13 % Mar The proportion of the meter reading period in March 0.00 .43 0.09 0.15 % Apr The proportion of the meter reading period in April 0.00 .43 0.09 0.13 % May The proportion of the meter reading period in May 0.00 .42 0.09 0.13 % Jun The proportion of the meter reading period in June 0.00 .42 0.09 0.14 % Jul The proportion of the meter reading period in July 0.00 .41 0.09 0.13 % Aug The proportion of the meter reading period in Aug 0.00 .41 0.09 0.13 % Sep The proportion of the meter reading period in September 0.00 .41 0.08 0.14 % Oct The proportion of the meter reading period in October 0.00 .43 0.08 0.13 % Nov The proportion of the meter reading period in November 0.00 .42 0.08 0.12 % Dec The proportion of the meter reading period in December 0.00 .42 0.07 0.13 % The number of observations for all variables is 28,473 and that there are no missing data for any of the variables listed above. Property size ranges from 100 to 1000 square meters, as these limitations were placed during the property selection process. It has a mean value of 576 square meters, with the standard deviation of 135 square meters, which indicates that there are a wide variety of properties in this analysis. The daily average water consumption is 0.65 kL (about 240 kL per year), which is roughly the same as the average water consumption across Sydney by each property. However, the standard deviation is
  • 53. 42 0.38 kL (almost 60% of the mean), which also shows that there is a huge variation in consumption. This was expected, as this study is looking at segments which are fundamentally different from each other. The average value of ‘Weighted Average Price’ falls between ‘Real Tier 1’ price and ‘Real Tier 2’ price. Tier 2 pricing is applicable when the annual consumption is more than 400 kL per year. However, the average consumption is about 240 kL per year. Hence, the ‘Weighted Average Price’ is a lot closer to ‘Real Tier 1’ price than ‘Real Tier 2’ price. Minimum values of weather variables seem a bit high and the maximum values of weather variables seem a bit low. This is because these observations are averaged over about ninety days, to match the metered water consumption period. As a result, there is very little variability observed on data on weather variables. Each month covers about 0.08% (1 month/12 months) of the period in a year. As expected, all monthly variables have average values which fall between 0.07 and 0.09. The average value for the restriction dummy variable also matches with the period water restriction Level 3 which was in place. In addition to the descriptive statistics indicates that the panel data is ‘strongly balanced’.
  • 54. 43 5.3 Consumer segments As pointed out earlier, 28,473 properties were selected for the analysis. Table 5-2 shows the exact number of properties that fall under each chosen consumer segment. Specification of each segment was earlier given in Section 4.1 (Residential consumer segments). Table 5-2: Consumer segments and number of properties Consumer Segment Number of properties (sample size)  Segment 01 ‐ Base  4,387    Segment 02 ‐ Townhouse  1,542    Segment 03 ‐ Gardeners  540    Segment 04 ‐ Pensioners  3,452    Segment 05 ‐ Swimming Pool  943    Segment 06 ‐ Financially Troubled  491    Segment 07 ‐ DoH Renters  294    Segment 08 ‐ Payment Plan  178    Segment 09 ‐ CLAD  666    Segment 10 ‐ Low Income  943    Segment 11 ‐ High Income  1,923    Segment 12 ‐ Water concise  2,068    Segment 13 ‐ Small lot size  1,774    Segment 14 ‐ Large lot size  763    Total  19,964    Each segment has reasonably enough number of properties to obtain sensible results from panel modelling. Around 8,500 properties were omitted as they were not mutually exclusive to any of the above chosen consumer segments. It is interesting to see how each segment’s water consumption varies during the analysis period. Figure 5-2 shows daily average water consumption observed during the period of analysis for each segment. However, it is the change in consumption that is of interest to this study. Base segment is very close to the average of all other consumer segments. It is worth noting that the base segment was not compiled based on water consumption; rather, it was determined based on social, economic and geographical characteristics.
  • 55. 44 Figure 5-2: Daily average water consumption within study period for each consumer segment Following from Figure 5-2, the water consumption of townhouses is considerably lower than the base segment which consists of larger houses. This result is as expected. Also, the data reveals that gardeners in this segment seem to be using an average amount of water. Even though pensioners are expected to use higher amount of water individually, pensioner households use less water than other segments that consist of houses. Swimming pool owners have the highest water use, which confirms with previous findings. Segments that are seemingly under financial stress appear to be using a lot more water than the average household. In contrast, renters use a lot less water than the base segment, as former only pay a portion of the total water bill. On the other hand, CALD households and water conscious use more water than the average household while the lower income households are using less water than the higher income households. Although there seems to be no difference between larger houses and smaller houses in regards to water consumption it is however important to note that both use less water than an average sized household.
  • 56. 45 Chapter 6 -M O D E L R E S U L T S A N D D I S C U S S I O N S 6.0 Introduction Fourteen separate models were devised representing each of the fourteen consumer segments identified in this study. This chapter explains in detail the steps employed to model the base segment. The selected base segment model specification is then fine-tuned to suit all of the consumer segments as the primary aim of this study is to compare the long-run price elasticity of demand for water of all the consumer segments. Figure 6-1 below illustrates how the analysis and discussion of this chapter is structured. Figure 6-1: Flow chart diagram mapping the steps taken to reach the final consumer segment models . Inadequate Adequate Best for the base segment Suitable for all segments
  • 57. 46 As argued in Section 4.3 (Estimation Methods), the dynamic nature of model specification is likely to produce biased estimates using the OLS, fixed effects or random effects estimators. Section 6.1 argues as to why the GMM specification is the best for this study after a thorough evaluation of the results of the commonly used estimators. Section 6.2 takes the initial GMM model specified in Section 6.1 and fine tunes it to suit all the fourteen consumer segments. Finally, Section 6.3 uses the final GMM model specifications to model each consumer segment separately. The modelling results are discussed for each consumer segment under different sub- sections. If the model specification changes due to the nature of the data set, then the changes are noted under each sub-section.

Related Documents