GEOGRAPHY AND THE COSTS OF URBAN ENERGY ...

GEOGRAPHY AND THE COSTS OF URBAN ENERGY INFRASTRUCTURE: THE CASE OF ELECTRICITY AND NATURAL GAS CAPITAL INVESTMENTS

Dissertation

Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University By Muzeyyen Anil Senyel, B.C.P. (High Hons.), M.C.P, M.C.R.P. Graduate Program in City and Regional Planning

The Ohio State University 2013

Dissertation Committee: Professor Jean-Michel Guldmann, Advisor Professor Rachel Garshick Kleit Professor Philip A. Viton

Copyright by Muzeyyen Anil Senyel 2013

ii

ABSTRACT

Investments in the urban energy infrastructure for distributing electricity and natural gas are analyzed using (1) property data measuring distribution plant value at the local/tax district level, and (2) system outputs such as sectoral numbers of customers and energy sales, input prices, company-specific characteristics such as average wages and load factor. Socio-economic and site-specific urban and geographic variables, however, often been neglected in past studies. The purpose of this research is to incorporate these site-specific characteristics of electricity and natural gas distribution into investment cost model estimations. These local characteristics include (1) socio-economic variables, such as income and wealth; (2) urban-related variables, such as density, land-use, street pattern, housing pattern; (3) geographic and environmental variables, such as soil, topography, and weather, and (4) company-specific characteristics such as average wages, and load factor. The classical output variables include residential and commercialindustrial customers and sales. In contrast to most previous research, only capital investments at the local level are considered. In addition to aggregate cost modeling, the analysis focuses on the investment costs for the system components: overhead conductors, underground conductors, conduits, poles, transformers, services, street lighting, and station equipment for electricity distribution; and mains, services, regular and industrial measurement and ii

regulation stations for natural gas distribution. The Box-Cox, log-log and additive models are compared to determine the best fitting cost functions. The Box-Cox form turns out to be superior to the other forms at the aggregate level and for network components. However, a linear additive form provides a better fit for end-user related components. The results show that, in addition to output variables and company-specific variables, various site-specific variables are statistically significant at the aggregate and disaggregate levels. Local electricity and natural gas distribution networks are characterized by a natural monopoly cost structure and economies of scale and density. The results provide evidence for the economies of scale and density for the aggregate electricity and natural gas distribution systems. However, distribution components have varying economic characteristics. The backbones of the networks (overhead conductors for electricity, and mains for natural gas) display economies of scale and density, but services in both systems and street lighting display diseconomies of scale and diseconomies of density. Finally multi-utility network cost analyses are presented for aggregate and disaggregate electricity and natural gas capital investments. Economies of scope analyses investigate whether providing electricity and natural gas jointly is economically advantageous, as compared to providing these products separately. Significant economies of scope are observed for both the total network and the underground capital investments.

iii

Dedicated to my parents

iv

ACKNOWLEDGMENTS

I would like to express my deepest and sincere appreciation to my advisor, Professor Jean-Michel Guldmann, for his guidance and encouragement throughout my graduate studies and completing this dissertation. His wise advices and critiques helped me progress in my research, and his positive attitude and kindness made me feel a valued student. He has always been an inspiration for me, not only in academic but also in personal aspects. I would like to thank Professors Philip A.Viton and Rachel Garshick Kleit for their precious time and comments, and for their much-admired professionalism. I am grateful for the scholarship provided by the Middle East Technical University, Turkey, and The Council of Higher Education, Turkey, which financed the first four years of my study. I would also like to thank the Department of City and Regional Planning at the Ohio State University for providing me teaching assistantships and instructor positions during the last two years of my study, which was an invaluable experience for me besides the financial support. It was a pleasure to work as a teaching assistant with Professors: Steven I. Gordon, Charisma Acey, Carolyn Merry, Kimberly Burton, and Abraham Ndungu. I extend my gratitude to the Professors in the Faculty of Science at Anadolu University, Turkey, for their help and insights in my research about statistics and mathematical methods. I would also thank to all my friends in the United States and in Turkey for their friendship and support. v

Finally, I want to recognize my family for their support. I owe many thanks to my parents, Fatma and Mustafa Senyel, for their patience, and unconditional love. They have always encouraged me to pursue my dreams.

vi

VITA January 8, 1981.…………………... Born – Eskisehir, Turkey 2003………………………………... B.C.P. (High Hons.), City and Regional Planning School of Architecture Middle East Technical University Ankara, Turkey. 2006………………………………... M.C.P., City Planning School of Architecture Middle East Technical University Ankara, Turkey. 2012………………………………... M.C.R.P., City and Regional Planning Austin E. Knowlton School of Architecture The Ohio State University Columbus, Ohio, U.S.A. 2005-present………………………... Graduate Research Associate Middle East Technical University Ankara, Turkey. 2011-present...……………………… Graduate Teaching Associate Austin E. Knowlton School of Architecture The Ohio State University Columbus, Ohio, U.S.A.

FIELDS OF STUDY Major Field: City and Regional Planning Minor Fields: Urban Energy Infrastructure Econometrics Geographical Information Systems vii

TABLE OF CONTENTS Page Abstract ............................................................................................................................... ii Acknowledgments................................................................................................................v Vita.................................................................................................................................... vii Table of Contents ............................................................................................................. viii List of Tables ................................................................................................................... xiii List of Figures ....................................................................................................................xx CHAPTERS CHAPTER 1: INTRODUCTION ........................................................................................1 CHAPTER 2: LITERATURE REVIEW .............................................................................6 2.1 Electricity Economics .............................................................................................. 6 2.1.1 Electricity Cost Models .................................................................................... 7 2.1.1.1 Modeling Multiple-Component Electricity Costs ................................ 8 2.1.1.2 Modeling Electricity Distribution Costs ............................................. 12 2.1.2 Modeling Electricity Distribution Efficiency ................................................. 18 2.1.2.1 Electricity Distribution Efficiency Analysis with the DEA Method .. 18 2.1.2.2 Electricity Distribution Efficiency Analyses with Parametric COLS and SFA Methods ........................................................................................... 23 2.1.2.3 Electricity Distribution Efficiency: Comparing Parametric and NonParametric Methods ........................................................................................ 27 2.1.2.4 Electricity Distribution Efficiency Studies Using Profit Function ..... 28 2.2 Natural Gas Economics .......................................................................................... 28 2.2.1 Natural Gas Distribution Cost Models ........................................................... 29 2.2.2 Natural Gas Distribution Efficiency Models .................................................. 33 2.2.2.1 Natural Gas Distribution Efficiency Analysis with the DEA Method 33 viii

2.2.2.2 Natural Gas Distribution Efficiency Analysis with the Parametric SFA Method............................................................................................................ 35 2.2.2.3 Natural Gas Distribution Efficiency Analysis with Profit and Cost Functions ........................................................................................................ 36 2.3 Economics of Other Utilities.................................................................................. 37 2.4 Economics of Multi-Sector Utilities ...................................................................... 39 2.5 Concluding Remarks .............................................................................................. 43 CHAPTER 3: METHODOLOGY .....................................................................................44 3.1 Introduction ............................................................................................................ 44 3.2 Cost Modeling in Neo-classical Economic Theory ............................................... 45 3.2.1 Total Cost Function ........................................................................................ 47 3.2.2 Capital Cost Function ..................................................................................... 48 3.2.3 Elasticity ......................................................................................................... 50 3.2.4 Economies of Scale and Density .................................................................... 51 3.2.5 Economies of Scope ....................................................................................... 52 CHAPTER 4: DATA SOURCES ......................................................................................54 4.1 Company Data ........................................................................................................ 54 4.1.1 Plant Investment Data .................................................................................... 55 4.1.2 Customers and Sales Data .............................................................................. 62 4.1.3 Employment Data ........................................................................................... 65 4.2 Census of Population ............................................................................................. 66 4.3 Economic Censuses ................................................................................................ 70 4.4 Geographic Data..................................................................................................... 72 4.4.1 Land-Use Data................................................................................................ 72 4.4.2 Slope Data ...................................................................................................... 76 4.4.3 Soil Data ......................................................................................................... 80 4.4.4 Street Data ...................................................................................................... 87 4.4.5 Meteorological Data ....................................................................................... 90 CHAPTER 5: MODELS OF AGGREGATE CAPITAL INVESTMENTS IN ELECTRICITY AND NATURAL GAS DISTRIBUTION ..............................................95 5.1 Aggregate Capital Investment Cost Functions ....................................................... 95 ix

5.1.1 Overview ........................................................................................................ 95 5.1.2 Cost Model for Aggregate Electricity Distribution Investment ..................... 99 5.1.2.1 Model Estimates ................................................................................. 99 5.1.2.2 Economies of Scale .......................................................................... 102 5.1.2.3 Economies of Density....................................................................... 108 5.1.3 Cost Model for Aggregate Natural Gas Distribution Investment................. 113 5.1.3.1 Model Estimation ............................................................................. 113 5.1.3.2 Economies of Scale .......................................................................... 117 5.1.3.3 Economies of Density....................................................................... 123 5.2 Multi-Utility Cost Analysis .................................................................................. 128 5.3 Summary .............................................................................................................. 134 CHAPTER 6: ANALYSES OF DISAGGREGATE CAPITAL INVESTMENTS IN ELECTRICITY AND NATURAL GAS DISTRIBUTION ............................................137 6.1 Disaggregate Cost Functions ............................................................................... 138 6.2 Cost Models for Electricity Distribution Network Components ......................... 139 6.2.1 Cost Model for Overhead Conductors ......................................................... 139 6.2.1.1 Cost Model for Primary Overhead Conductors ................................ 143 6.2.1.2 Cost Model for Secondary Overhead Conductors ............................ 146 6.2.2 Cost Model for Underground Conductors .................................................... 148 6.2.2.1 Cost Model for Primary Underground Conductors .......................... 152 6.2.2.2 Cost Model for Secondary Underground Conductors ...................... 155 6.2.3 Cost Model for Conduits .............................................................................. 158 6.2.4 Cost Model for Services ............................................................................... 164 6.2.5 Cost Model for Poles .................................................................................... 167 6.2.6 Cost Model for Transformers ....................................................................... 170 6.2.7 Cost Model for Street Lighting .................................................................... 173 6.2.8 Cost Model for Station Equipment............................................................... 175 6.3 Cost Models for Natural Gas Distribution Network Components ....................... 178 6.3.1 Cost Model for Distribution Mains .............................................................. 178 6.3.2 Cost Model for Gas Services ........................................................................ 181 x

6.3.3 Cost Model for Measurement and Regulation Stations................................ 185 6.3.4 Cost Model for Industrial Measurement and Regulation Stations ............... 189 6.4 Marginal Costs and Elasticities in Total Capital Investment ............................... 191 6.4.1 Total Capital Investment: Electricity ........................................................... 192 6.4.2 Marginal Capital Costs: Electricity .............................................................. 194 6.4.3 Economies of Scale: Electricity ................................................................... 197 6.4.4 Total Capital Investment: Natural Gas ......................................................... 198 6.4.5 Marginal Capital Costs: Natural Gas ........................................................... 200 6.4.6 Economies of Scale: Natural Gas ................................................................. 202 6.5 Multi-Utility Cost Analysis of Underground Network Investments .................... 203 6.6 Summary .............................................................................................................. 210 CHAPTER 7: CONCLUSIONS ......................................................................................215 BIBLIOGRAPHY ............................................................................................................222 APPENDIX APPENDIX A: MODELS OF ELECTRICITY AND NATURAL GAS CUSTOMERS AND SALES ....................................................................................................................233 A.1 Introduction ......................................................................................................... 234 A.2 Estimation Models .............................................................................................. 237 A.2.1 Electricity Models ....................................................................................... 237 A.2.1.1 Modeling the Number of Residential Electricity Customers .......... 237 A.2.1.2 Modeling Residential Electricity Sales............................................ 239 A.2.1.3 Modeling the Number of Commercial-Industrial Electricity Customers ..................................................................................................... 241 A.2.1.4 Modeling Commercial-Industrial Electricity Sales ......................... 242 A.2.1.5 Modeling Lighting Customers and Sales ........................................ 243 A.2.2 Natural Gas Models ..................................................................................... 245 A.2.2.1 Modeling the Number of Residential Natural Gas Customers ........ 245 A.2.2.2 Modeling Residential Natural Gas Sales ......................................... 247 A.2.2.3 Modeling Commercial-Industry Natural Gas Customers ................ 248 A.2.2.4 Modeling Commercial-Industrial Natural Gas Sales ...................... 248 xi

A.3 Explanation Models............................................................................................. 249 A.3.1 Electricity Models ....................................................................................... 249 A.3.1.1 Modeling Residential Electricity Sales............................................ 249 A.3.1.2 Modeling Commercial-Industrial Electricity Customers ................ 250 A.3.1.3 Modeling Commercial-Industrial Electricity Sales ......................... 252 A.3.2 Natural Gas Models ..................................................................................... 253 A.3.2.1 Modeling Residential Natural Gas Sales ......................................... 253 A.3.2.2 Modeling Commercial-Industrial Natural Gas Customers .............. 254 A.3.2.3 Modeling Commercial-Industrial Natural Gas Sales ...................... 255 A.4 Concluding Remarks ........................................................................................... 255 APPENDIX B: ECONOMIES OF SCALE IN AGGREGATE CAPITAL INVESTMENT ................................................................................................................257 B.1 Economies of Scale in Aggregate Electricity Capital Investment....................... 257 B.2 Economies of Scale in Aggregate Natural Gas Capital Investment .................... 260 APPENDIX C: SCOPE SCORES FOR DIFFERENT LEVELS OF SITE-SPECIFIC VARIABLES AND DIFFERENT OUTPUT COMBINATIONS ..................................264 C.1 Scope Score Tables for Aggregate Multi-Utility Costs....................................... 264 C.2 Scope Score Tables for Underground Multi-Utility Costs .................................. 266

xii

LIST OF TABLES Table

Page

Table 2.1 Multiple-component Studies of the Electric Industry ........................................10 Table 2.2 Distribution Cost Studies in the Electricity Industry .........................................13 Table 2.3 Efficiency Modeling of Electricity Utilities Using DEA...................................20 Table 2.4 Efficiency Studies on Electricity Utilities Using COLS and SFA ....................24 Table 2.5 Efficiency Studies of Electricity Utilities: Comparing Different Methods .......27 Table 2.6 Distribution Cost Studies in the Gas Industry ...................................................30 Table 2.7 Efficiency Studies on Natural Gas Utilities Using Non- Parametric Methods (DEA) .............................................................................................................34 Table 2.8 Efficiency Studies on Natural Gas Utilities Using Parametric Methods (SFA) 36 Table 2.9 Cost Modeling for Water and Telephone Utilities.............................................39 Table 2.10 Multiple-sector Utilities Cost Modeling ..........................................................40 Table 4.1 Descriptive Statistics for the Costs of the Historical Electricity Distribution Plant ($) ..........................................................................................................56 Table 4.2 Electricity Distribution Plant Accounts and Their Shares of the Total Cost .....58 Table 4.3 Descriptive Statistics for the Historical Natural Gas Distribution Plant ($) ......60 Table 4.4 Natural Gas Distribution Plant Accounts and Their Shares of the Total Cost ..61 Table 4.5 Descriptive Statistics for Electricity Customers and Sales ................................64 Table 4.6 Descriptive Statistics for Natural Gas Customers and Sales .............................64 Table 4.7 Company Electricity and Natural Gas Load Factors .........................................65 Table 4.8 Company Employees and Average Wages ($) ..................................................66 Table 4.9 Descriptive Statistics for Selected Census Variables ........................................68 Table 4.10 Descriptive Statistics for Selected Census Variables for the Tax Districts with Sales and Customers Data ......................................................................69 xiii

Table 4.11 Descriptive Statistics for Economic Census Sectors .......................................71 Table 4.12 Descriptive Statistics for Economic Census Sectors for the Tax Districts with Sales and Customers Data ......................................................................72 Table 4.13 Descriptive Statistics for Land-use Types (square miles) ...............................74 Table 4.14 Descriptive Statistics for Land-use Types (square miles) for the Tax Districts with Sales and Customers Data........................................................75 Table 4.15 Slope Groups and Slope Ranges ......................................................................78 Table 4.16 Descriptive Statistics for Areas with Different Slope Types (square miles) ...80 Table 4.17 Descriptive Statistics for Areas with Different Slope Types (square miles) for the Tax Districts with Sales and Customers Data .....................................80 Table 4.18 Descriptive Statistics for Soil Workability ......................................................83 Table 4.19 Descriptive Statistics for Soil Workability for the Tax Districts with Sales and Customers Data ........................................................................................83 Table 4.20 Descriptive Statistics for Steel Corrosion and Concrete Corrosion .................85 Table 4.21 Descriptive Statistics for Steel Corrosion and Concrete Corrosion for the Tax Districts with Sales and Customers Data ................................................85 Table 4.22 Descriptive Statistics for Average Rock Depth and Water Table ...................86 Table 4.23 Descriptive Statistics for Average Rock Depth and Water Table for the Tax Districts with Sales and Customers Data........................................................86 Table 4.24 Descriptive Statistics for Street Types (miles) ................................................89 Table 4.25 Descriptive Statistics for Street Types (miles) for the Tax Districts with Sales and Customers Data ..............................................................................89 Table 4.26 Stations and Statistics on Tax Districts within their Catchment Areas ...........91 Table 4.27 Descriptive Statistics for Meteorological Data ................................................92 Table 4.28 Descriptive Statistics for Meteorological Data for the Tax Districts with Sales and Customers Data ..............................................................................93 Table 5.1 Descriptive Statistics for the Aggregate Electricity Distribution Investment Cost Model (n=241) .....................................................................................100 Table 5.2 Aggregate Electricity Distribution Cost Function Estimates (n=241) .............101

xiv

Table 5.3 Aggregate Electric Distribution Cost Elasticities at the Sample Mean (n=241) .........................................................................................................102 Table 5.4 Descriptive Statistics for the Total Natural Gas Investment Cost Model (n=190) .........................................................................................................115 Table 5.5 Total Natural Gas Distribution Cost Function Estimates (n=190) ..................116 Table 5.6 Aggregate Natural Gas Distribution Cost Elasticities (n=190) .......................116 Table 5.7 Statistics for the Tax Districts with Diseconomies of Scale: εCG>1 ................118 Table 5.8 Descriptive Statistics for the Multi-utility Investment Cost Model (n=246) ...129 Table 5.9 Multi-utility Cost Function (n=246) ................................................................130 Table 5.10 Scope Scores for Different Levels of Electricity and Natural Gas Outputs ..131 Table 5.11 Scope Values for Different Levels of Site-specific Variables and Different Output Combinations....................................................................................132 Table 6.1 Descriptive Statistics for the Overhead Conductor Investment Cost Model (n=241) .........................................................................................................141 Table 6.2 Overhead Conductors Cost Function Estimates (n=241) ................................142 Table 6.3 Overhead Conductor Cost Elasticities at the Sample Mean (n=241) ..............142 Table 6.4 Descriptive Statistics for the Primary Overhead Conductor Investment Cost Model (n=137) ..............................................................................................144 Table 6.5 Primary Overhead Conductors Cost Function Estimates (n=137) ..................145 Table 6.6 Primary Overhead Conductor Cost Elasticities at the Sample Mean (n=137) 145 Table 6.7 Descriptive Statistics for the Secondary Overhead Conductor Investment Cost Model (n=137) ..............................................................................................147 Table 6.8 Secondary Overhead Conductors Cost Function Estimates (n=137)...............147 Table 6.9 Secondary Overhead Conductor Cost Elasticities at the Sample Mean (n=137) .........................................................................................................148 Table 6.10 Descriptive Statistics for the Underground Conductor Investment Cost Model (n=225) ..............................................................................................149 Table 6.11 Underground Conductors Cost Function Estimates (n=225) .........................150 Table 6.12 Underground Conductor Cost Elasticities at the Sample Mean (n=225) ......151

xv

Table 6.13 Descriptive Statistics for the Primary Underground Conductor Investment Cost Model (n=124) .....................................................................................153 Table 6.14 Primary Underground Conductors Cost Function Estimates (n=124) ...........154 Table 6.15 Primary Underground Conductor Cost Elasticities at the Sample Mean (n=124) .........................................................................................................155 Table 6.16 Descriptive Statistics for the Secondary Underground Conductor Investment Cost Model (n=123) .....................................................................................156 Table 6.17 Secondary Underground Conductors Cost Function Estimates (n=123) .......157 Table 6.18 Secondary Underground Conductor Cost Elasticities at the Sample Mean (n=123) .........................................................................................................157 Table 6.19 Descriptive Statistics for the Conduit Investment Cost Model Alternative 1 (n=434) .........................................................................................................159 Table 6.20 Conduit Cost Function Estimates, Model Alternative 1 (n=434) ..................160 Table 6.21 Conduit Cost Elasticities at the Sample Mean, Model Alternative 1 (n=434) .........................................................................................................161 Table 6.22 Descriptive Statistics for the Conduit Investment Cost Model Alternative 2 (n=122) .........................................................................................................162 Table 6.23 Conduit Cost Function Estimates, Model Alternative 2 (n=122) ..................163 Table 6.24 Conduit Cost Elasticities at the Sample Mean, Model Alternative 2 (n=122) .........................................................................................................163 Table 6.25 Descriptive Statistics for the Service Investment Cost Model (n=137).........165 Table 6.26 Service Cost Function Estimates (n=137) .....................................................166 Table 6.27 Service Cost Elasticities at the Sample Mean (n=137) ..................................167 Table 6.28 Descriptive Statistics for the Pole Investment Cost Model (n=137) .............168 Table 6.29 Pole Cost Function Estimates (n=137) ..........................................................169 Table 6.30 Pole Cost Elasticities at the Sample Mean (n=137) .......................................170 Table 6.31 Descriptive Statistics for the Transformer Investment Cost Model (n=225) 171 Table 6.32 Transformer Cost Function Estimates (n=225) .............................................172 Table 6.33 Pole Cost Elasticities at the Sample Mean (n=225) .......................................172

xvi

Table 6.34 Descriptive Statistics for the Street Lighting Investment Cost Model (n=233) .........................................................................................................173 Table 6.35 Street Lighting Cost Function Estimates (n=233) .........................................174 Table 6.36 Street Lighting Cost Elasticities at the Sample Mean (n=233) ......................175 Table 6.37 Descriptive Statistics for the Station Equipment Investment Cost Model (n=133) .........................................................................................................176 Table 6.38 Station Equipment Cost Function Estimates (n=133) ...................................177 Table 6.39 Station Equipment Cost Elasticities at the Sample Mean (n=233) ................177 Table 6.40 Descriptive Statistics for the Distribution Mains Investment Cost Model (n=190) .........................................................................................................179 Table 6.41 Distribution Mains Cost Function Estimates (n=190) ...................................180 Table 6.42 Distribution Mains Cost Elasticities at the Sample Mean (n=190) ...............181 Table 6.43 Descriptive Statistics for the Service Investment Cost Model (n=190).........182 Table 6.44 Service Cost Function Estimates (n=190) .....................................................183 Table 6.45 Service Cost Elasticities at the Sample Mean (n=190) ..................................185 Table 6.46 Descriptive Statistics for the Service Investment Cost Model (n=135).........186 Table 6.47 Measurement and Regulation Station Cost Function Estimates (n=135) ......187 Table 6.48 Measurement and Regulation Stations Cost Elasticities at the Sample Mean (n=135) .........................................................................................................188 Table 6.49 Descriptive Statistics for the Industrial Measurement and Regulation Stations Cost Model (n=41) .........................................................................189 Table 6.50 Industrial Measurement and Regulation Stations Cost Function Estimates (n=41) ...........................................................................................................190 Table 6.51 Industrial Measurement and Regulation Stations Cost Elasticities at the Sample mean (n=41) ....................................................................................191 Table 6.52 Characteristics of the Electricity Sample (n=125) .........................................193 Table 6.53 Estimated Components and Total Capital Costs............................................194 Table 6.54 Estimated Component Marginal Capital Costs at the Electricity Sample Mean ($) .......................................................................................................197

xvii

Table 6.55 Elasticities, and Ray Economies of Scale and Density at the Electricity Sample Mean (n=125) ..................................................................................198 Table 6.56 Characteristics of the Natural Gas Sample (n=33) ........................................199 Table 6.57 Estimated Components and Total Capital Costs............................................200 Table 6.58 Estimated Component Marginal Capital Costs for the Natural Gas Sample ($) ....................................................................................................201 Table 6.59 Elasticity, and Ray Economies of Scale and Density in Natural Gas Sample Mean (n=33) .................................................................................................202 Table 6.60 Descriptive statistics for the underground multi-utility investment cost model (n=234) ..............................................................................................204 Table 6.61 Multi-utility underground cost function estimates (n=234) ...........................205 Table 6.62 Scope values for different levels of electricity and natural gas outputs ........206 Table 6.63 Scope values for different levels of site-specific variables and different output combinations. ....................................................................................207 Table A.1 Correlations among the Houses Using Electricity for Space Heating (HHEATE), Cooking (HCOOKE), and Water Heating (WHEATE) ........................239 Table A.2 Correlations among the Number of Establishments: Manufacturing (MAEST), Retail (REEST), Services (SEEST), Wholesale (WSEST) .................................251 Table A.3 Correlations among the Number of Employees: Manufacturing (MAEMP), Retail (REEMP), Services (SEEMP), Wholesale (WSEMP) ...............................252 Table C.1 Scope Scores for the Minimum Share of 40+ Year Old Houses (AGE_P40=0.003) .........................................................................................264 Table C.2 Scope Scores for the Maximum Share of 40+ Year Old Houses (AGE_P40=0.90) ...........................................................................................265 Table C.3 Scope Scores for the Minimum Number of Street Intersections (INT=13) ....265 Table C.4 Scope Scores for the Maximum Number of Street Intersections (INT=15,114) ...............................................................................................266 Table C.5 Scope Scores for the Minimum Built-up Area (ABLTP=0.04) .........................266 Table C.6 Scope Scores for the Maximum Built-up Area (ABLTP=133.85) .....................267 Table C.7 Scope Scores for the Minimum Soil Corrosivity (SOILCORR=0.03) ...............267 xviii

Table C.8 Scope Scores for the Maximum Soil Corrosivity (SOILCORR=0.98) ..............268 Table C.9 Scope Scores for the Minimum Share of Buildings with 1 to 3 Stories (STO_P1to3=0.64)..........................................................................................268 Table C.10 Scope Scores for the Maximum Share of Buildings with 1 to 3 Stories (STO_P1to3=0.99)..........................................................................................269 Table C.11 Scope Scores for the Minimum Average Heating Degree Days (HDD=454) ..................................................................................................269 Table C.11 Scope Scores for the Maximum Average Heating Degree Days (HDD=731) ..................................................................................................270

xix

LIST OF FIGURES Figure

Page

Figure 4.1 Geographic Distribution of the Tax Districts of the Selected Electricity Companies ......................................................................................................56 Figure 4.2 Geographic Distribution of Electricity Distribution Total Costs ($1,000) .......57 Figure 4.3 Geographic Distribution of the Tax Districts of the Selected Natural Gas Provider Companies .......................................................................................59 Figure 4.4 Distribution of Natural Gas Distribution Plant Total Costs ($1,000) ...............60 Figure 4.5 Geographic Distribution of Tax Districts with Electricity Sales and Customers Data ..............................................................................................62 Figure 4.6 Geographic Distribution of Tax Districts with Natural Gas Sales and Customers Data ..............................................................................................63 Figure 4.7 Population Distribution in the State of New York ...........................................70 Figure 4.8 Tax Districts with Economic Censuses ............................................................71 Figure 4.9 Map of Land Use Categories in New York State .............................................73 Figure 4.10 Topography Map of New York State .............................................................77 Figure 4.11 Relationship between Slope in Degrees and Slope in Percent. ......................79 Figure 4.12 Soil Types in New York State (there are 169 soil types coded with regard to MUID grids, which are re-grouped according to the workability, corrosivity, average rock depth and water table characteristics, in the analysis) ..........................................................................................................81 Figure 4.13 Major Street Types in New York State ..........................................................88 Figure 4.14 Meteorological Stations and Their Catchment Areas .....................................90 Figure 4.15 Variations of Annual Heating Degree Days across the State of New York ...94 Figure 4.16 Variations of Annual Cooling Degree Days across the State of New York ...94 xx

Figure 5.1 Point Elasticities of Scale across 241 Observations (tax districts). ................103 Figure 5.2 Economies of Scale, εCE, versus Output Parameter, k. ...................................105 Figure 5.3 Economies of Scale, εCE, versus Output Parameter, k, for Different Levels of the Site-specific, Geographic, and Company-specific Variables: (a) Area density, (b) Soil corrosivity, (c) Number of street intersections, (d) Electricity load factor ...................................................................................106 Figure 5.4 Point Elasticities of Density across 241 Observations (tax districts). ............109 Figure 5.5 Economies of Density, εDE, versus Output Parameter, k. ...............................110 Figure 5.6 Economies of Density, εDE, versus Output Parameter, k, for Different Levels of the Site-specific, Geographic, and Company-specific Variables: (a) Area density, (b) Soil corrosivity, (c) Number of street intersections, (d) Electricity load factor ...................................................................................111 Figure 5.7 Point Elasticities of Scale across 190 Observations (tax districts). ................118 Figure 5.8 Economies of Scale, εCG, versus Output Parameter, k. ...................................119 Figure 5.9 Economies of Scale, εCG, versus Output Parameter, k, at Different Levels of the Site-specific, Geographic, and Company-specific Variables: (a) Builtup area, (b) Soil workability, (c) Number of street intersections, (d) Wage per employee. ...............................................................................................121 Figure 5.10 Point Elasticities of Density across 190 Observations (tax districts). ..........124 Figure 5.11 Economies of Density, εDG, versus Output Parameter, k. .............................125 Figure 5.12 Economies of Density, εD, versus Output Parameter, k, at Different Levels of the Site-specific, Geographic, and Company-specific Variables: (a) Built-up area, (b) Soil workability, (c) Number of street intersections, (d) Wage per employee. .....................................................................................126 Figure 5.13 Scope Scores for Different Levels of Electricity and Natural Gas Outputs. 132 Figure 5.14 Scope Scores for Different Values of Electricity and Natural Gas Sales, and for Different Levels of Site-specific, Geographic, and Company-specific Variables: (a) Number of street intersections, (b) Share of 40+ year old houses ...........................................................................................................133 Figure 6.1 Scope scores for different levels of electricity and natural gas outputs. ........206 xxi

Figure 6.2 Scope scores for different values of electricity and natural gas sales, and for different levels of site-specific, geographic, and company-specific variables: (a) Built-up area, (b) Soil corrosivity, (c) Average heating degree days, (d) Share of buildings with 1 to 3 stories ............................................208 Figure A.1 Distribution of NRE and HH ...........................................................................238 Figure A.2 Distribution of SRE .........................................................................................240 Figure A.3 Distribution of NL and SL ..............................................................................244 Figure A.4 Distribution of NRG and SRG ..........................................................................246 Figure B.1 Constant Returns to Scale Tradeoff Curve ( Communities where

) for the Three

: (a). Havestraw Village, (b) Wappingers

Falls Village, (c) Grandview on Hudson ......................................................262

xxii

CHAPTER 1

INTRODUCTION

The urban energy infrastructure is composed of electricity and natural gas distribution systems. Energy provision is fundamental to sustaining daily life and industrial production. Therefore, urban energy networks are closely related to the socioeconomic and site-specific characteristics of the service area. Energy distribution infrastructure investments follow and also lead urban development. This research aims to expand the existing research literature on electricity and natural gas cost analysis by considering socio-economic, site-specific, and company-specific characteristics. Before discussing the economics of the energy infrastructure, it is important to understand the characteristics of the electricity and natural gas industries. Electricity is a secondary source of energy, converted from primary energy sources such as coal, nuclear or renewable energy. Electricity provision includes three stages: generation, transmission and distribution. The most widely used production systems in the U.S. are steam turbines, which are powered by coal, natural gas or oil. After electricity is generated in power plants, transformers step up voltage for transmission. High voltage is necessary to minimize energy losses during transfer. High voltage electricity is carried by transmission lines over long distances. The voltage is 1

stepped down by substations and transformers before entering the local network and reaching final consumers. The final voltage reduction occurs at transformers, generally attached to poles, before reaching the end users. The system is designed to meet peak load requirements and provide a reliable service at a given voltage level while considering safety and environmental issues. The U.S. market included, in 2007, more than 3200 traditional electric utilities, which provide adequate and reliable electricity to consumers in their service territories. Electric utilities can be investor-owned, publicly-owned, Federal utilities or cooperatives. Power marketers also buy and sell electricity, but usually do not own or operate generation, transmission, or distribution facilities (EIA, Electricity). The electricity industry has gone through several federal regulations changes since the 1970s. The Public Utilities Regulatory Policies Act of 1978 has allowed non-utility generators to provide independent power. The Energy Policy Act of 1992 promoted competition in wholesale electricity markets by opening access to transmission networks (Nunez, 2007:195). Regulations of interstate sales of electricity are the purview of the Federal Energy Regulatory Commission (FERC), while interstate sales are the purview of state regulatory commissions. Natural gas is a primary form of energy, used mostly for heating or electricity production. The gas industry includes four stages: extraction, production, transmission and distribution. Natural gas can be stored for later use, unlike electricity. Gas is extracted from oil and gas wells, and is next processed to separate it from liquids. The dry gas is then transmitted at high pressures over long pipelines. This pressure is stepped down at city-gate stations. Finally, gas is distributed to end users through distribution 2

mains and services. The U.S. natural gas transportation system includes mainline natural gas pipeline companies and local distribution companies (LDCs). LDCs can be investorowned, privately owned, municipal or cooperatives. “The plant structure of gas distribution utilities may vary significantly from one company to another, depending upon supply conditions, local market factors, and historical circumstances” (Guldmann, 1983:300). The number of customers and the amount of natural gas supplied define market size. The Natural Gas Policy Act (1978) removed price ceilings on production, which had resulted in imbalance in supply and demand, and gas shortages in the 1970s. The Federal Energy Regulatory Commission issued two orders in 1985 and 1992, which created a more flexible and competitive market, including open-access transportation for large-volume customers (EIA, Distribution of Natural Gas, 2008). The gas distribution system is designed to provide adequate supply at proper pressure to customers. System upgrade through reinforcement, pressuring and extension may be needed. Gas utilities, similar to electricity ones, try to minimize costs under regulated prices for distribution. The presence of an energy substitute is another issue that natural gas utilities must take into account, as customers may shift to an alternative resource if the use of natural gas becomes economically infeasible. These economic considerations encourage utilities to estimate their current and future costs in the most accurate way, and apply efficient pricing mechanisms. Energy infrastructure investments are close to irreversible. Once their construction is completed, it is very difficult and costly to modify these systems, particularly if underground. Their cost structure is a critical component in the decision3

making of gas and electric utilities. If realistic cost functions can be derived for both industries, resources can be allocated in a more efficient way, while providing safe and reliable energy. Electricity and natural gas distribution cost structures have been analyzed in the literature, with determinants including output variables and a limited number of environmental characteristics such as density. However, there is a complex interaction between distribution systems and the socio-economic and site-specific characteristics of the service area. Using a more detailed set of explanatory variables, such as socioeconomic characteristics, land use pattern, soil, slope, weather, and company-specific characteristics such as load factor and average wages, should lead to more comprehensive investment costs functions that can be used for cost forecasting and the assessment of the economic feasibility of infrastructure expansion plans. In addition to these cost function determinants, economies of scale, density and scope are also important issues in energy infrastructure economics. An assessment of these economies is critical for decision making and public policy regarding competition at the local level. The basic research question that needs an answer is: Are there specific local urban and environmental conditions that would lead to diseconomies of scale, density, and scope, and therefore would be conducive to competition, with multiple utilities operating at the local level? A better understanding of the structure of energy infrastructure investments is becoming even more important in the light of emerging technologies, such as the smart grid in electricity, which will enable on-site energy production and the integration of renewable energies, thus empowering consumers. The traditional grid has not been

4

abandoned, and still dominates the industry. However, such a better understanding will help manage future investments under new technologies and market conditions. The existing research literature generally deals with company-level aggregate data and few environmental variables, using a variety of functional forms. This literature is reviewed in Chapter 2. This research, in contrast, deals with the local-level data of four electricity and natural gas utilities in the State of New York and expands the existing research literature by considering the (1) socio-economic characteristics of the service territories, such as population and income, (2) site-specific housing characteristics and urban pattern such as density, house size, age of the housing stock, and street pattern, (3) geographic characteristics such as soil types, and topography, (4) environmental factors such as weather, and (5) company-specific factors such as average wages and load factors. Modeling approaches are discussed in Chapter 3. The tax district is the unit area in this research, and varies between 3 and 483 square miles. Although all socio-economic and geographic variables are available for all the 1619 tax districts of the State of New York, company market data are more limited. Details on these data are presented in Chapter 4. Chapter 5 presents economic analyses of aggregate-level capital investments. Chapter 6 presents similar analyses at the disaggregate level. Finally, conclusions are presented in Chapter 7.

5

CHAPTER 2

LITERATURE REVIEW

Studies on the cost structure of urban energy distribution utilities – electricity and natural gas – date back to the 1970s. Early research focused on the whole industry, combining the generation, transmission, and distribution components. Research on the monopolistic structure of the industry and scale economies then shifted to of each component separately. Efficiency in electricity and natural gas systems became a focus in the 1990s, with benchmarking techniques used to design policies and regulations. Market characteristics, such as numbers of customers and sales, were considered in all these studies. However, urban, geographic and environmental factors were often neglected. Further, research on multi-utility network systems and economies of scope has been quite limited. The purpose of this literature review is to critically analyze these different research streams.

2.1 Electricity Economics

The economics of electricity distribution focus on cost and efficiency modeling. Electricity distribution costs are discussed both in combination with generation and/or 6

transmission, and as stand-alone costs. Although multiple-component models are not directly related to the central theme of this research, they provide insights into system characteristics. Parametric and non-parametric efficiency models also inform the structure of distribution costs.

2.1.1 Electricity Cost Models

Electric utilities aim to minimize their costs while satisfying community needs for reliable and fairly priced service. Cost models explain the factors affecting the system. Guldmann (1988) states that a large amount of research has focused primarily on the economics of power generation, such as optimization of plant mix and capacity expansion, econometric analysis of scale effects, etc. Among prominent studies of the economics of power generation are Christiensen and Greene’s (1976) research on economies of scale in electricity generation, Schmalensee and Joskow’s (1986) estimation of the costs of electric generating units, Mazumdar and Kapoor’s (1995) study of power generation system production costs, Porat et. al.’s (1997) analysis of long-run electricity generation costs, and Maloney’s (2001) investigation of electricity generation costs. Electricity distribution economics, on the other hand, have not been studied as extensively as power generation. “Little theoretical or empirical work focuses explicitly on the economic characteristics of electric power distribution systems” (Joskow and Schmalensee, 1983: 59). Although the economics of distribution networks were considered of secondary importance, these networks constitute almost 30% of the total 7

investment in the electricity industry, and are directly related to public policy issues such as service provision, cost allocation and fair pricing. “The cost of electricity is a vital input to the price of almost all products and services” (Ingco, 1996:723). Therefore, further analyses of distribution costs should lead to better use of resources, more effective public policies, and a more efficient pricing mechanism. The literature on electricity distribution cost modeling includes two streams: (1) modeling multiple system components, including distribution, and (2) modeling distribution only.

2.1.1.1 Modeling Multiple-Component Electricity Costs

Early electricity economics research focuses on the whole system rather than merely on distribution (Table 2.1). Primeaux (1975) models the costs of electric utilities to examine their monopolistic structure, but does not distinguish among components, making it impossible to infer the economics of electricity distribution. Primeaux suggests that a policy shift from a monopolistic structure to a competitive one might provide social benefits. Weiss (1975) and Nelson and Primeaux (1988) examine the combined costs of transmission and distribution. Weiss (1975) models generation and transmissiondistribution costs, but his models have been criticized because of their simple linear form, with two explanatory variables – the number of customers and the load factor – while omitting several potentially relevant variables. Nelson and Primeaux (1988) examine transmission-distribution costs with a focus on monopoly structure. When competition is 8

eliminated in the market and only one firm remains as monopolist, the average total cost declines with an increase in output and number of customers for that firm. However when competition is introduced in a monopolistic market, the average cost per customer or sales is reduced, due to the decrease in output and number of customers. The results favor competition in larger markets, particularly after exhausting economies of density. Meyer (1975) models the cost of each component separately: production, transmission, distribution, and maintenance. The variables in the transmission and distribution models are the same, with different significance levels. He also compares public and private firms, and concludes that the average costs of private firms are higher than those of municipal firms, except for the large commercial customers. Huettner and Landon (1978) estimate production, transmission, distribution, fixed investment and other costs separately, in a way similar to Meyer. This is the first study examining the effects of regional differences and various customer groups by using regional dummy variables. However, these variables are not statistically significant. The empirical evidence supports economies of scale up to the average size and diseconomies of scale beyond, except for one operating cost category. Density is defined as the number of line transformers per customers. The authors assume that the expected sign for the density variable is positive, but the results contradict this assumption. In fact, the effect of density is ambivalent. The negative sign of density is plausible, since decreasing density implies a dispersed urban pattern, which is expected to increase plant investment costs. Higher density may decrease costs because of scale economies, but may increase them because of congestion.

9

Author(s) and Year Primeaux

Dependent Variable(s) Cost

(1975)

(generation+

fuel costs, purchased power,

224 observations

transmission+

consumption/customers, market density factor,

5-year time series

distribution)

Independent Variables Sales, generating capacity,

Functional form Linear

Study Area and Sample Size US cross section and

internal combustion dummies (generation, state, competition) Weiss

Cost/sales

Number of customers,

(1975)

(transmission+

load density

Linear

US 30 utilities

distribution)

cross-section

Meyer

Cost


(1975)

(production,

percent resale,

transmission,

public/private dummy,

distribution,

generation,

maintenance)

generation squared,

Quadratic

US 30 utilities 3-year panel (90 observations) cross-section pooled

generation cubed

(60 observations)

Huettner

Cost

Total capacity, total capacity

Quadratic

US

and

(production,

squared, utilization of

74 utilities

Landon

transmission,

capacity, utilization of

cross-section

(1978)

distribution, fixed investments)

capacity squared, regional dummies (northeast, north-central, west), wages, line transformers/customers, sales/customer group (residential, commercial, industrial), distributor dummies

Roberts

Cost


(1986)

(production+

sales (high, low voltages)

delivery)

prices (input, labor, ,

Translog

US 65 utilities cross-section

transmission, distribution) Nelson

Cost


and

(transmission+

length of transmission line,

Primeaux

distribution)

sales, price of electricity,

(1988)

Cobb-Douglas Quadratic

US 23 firms 16-year panel

wage, time trend, area, competition dummy

Continued Table 2.1 Multiple-component Studies of the Electric Industry 10

Table 2.1 continued Kaserman

Cost

Generation and distribution

Multistage

US

and Mayo

(generation+

amounts, customer shares

quadratic,

71 firms

(1991)

distribution)

(residential, industrial), prices

Flexible fixed

cross-section

(fuel, labor, capital, power)

costs,

generation capital (nuclear,

quadratic

hydro, gas), underground conduit-conductors-devices share, dummies (ownership, region, generation-distribution) Gilsdorf

Cost

Generation amount, sales,

(1995)

(generation+

labor, fuel, capital, utilization,

Translog

US 72 firms

distribution)

density, sales mix

Kwoka

Cost


cross-section

(1996)

(generation+

amounts, generation fixed

543

distribution)

costs, competition, power

cross-section

Quadratic

US

pools, membership, region, combined utilities, incentive regulation, capacity (nuclear, hydro, other), high voltage power, price (fuel, labor, capital), customers (residential, industrial, commercial) Thompson

Cost

Number of customers, sales

(1997)

(procurement

(high, low voltages), area, time,

+delivery)

power price, labor and capital

5-year pooled

prices (transmission, distribution)

cross-section

Kwoka

Cost


(2002)

(generation+

amounts, prices (labor, capital,

distribution)

fuel), shares of generation

Translog

US 83 firms

Quadratic

US 147 firms cross-section

capacity (nuclear, hydro), capacity utilization, high voltage power, density, average usage (residential, commercial, industrial) Fraquelli

Cost


et al.

(generation+

amounts, price (labor, other),

(2005)

distribution)

cost shares (labor, other),

7-year panel,

density

cross-section

11

Quadratic

Italy 25 utilities

Henderson (1985) analyzes the economics of vertically integrated electricity firms using different generation technologies. Distribution costs are not the primary concern of the study. However, the results show that generation and distribution cannot be functionally separated. Roberts (1986), Kaserman and Mayo (1991), Gilsdorf (1995), Thompson (1997), and Kwoka (1996, 2002) examine the whole industry: production, transmission and distribution. Their findings are in accordance with those of Henderson: vertical integration and inseparability of the system. Roberts assumes the possibility of cost substitution among the different components. Kaserman and Mayo, and Kwoka identify cost savings from vertical integration, except for small firms. Gilsdorf (1995), in contrast, does not observe subadditivity conditions for vertically-integrated electricity utilities, and refutes the hypotheses of natural multiproduct monopoly. More recently, Fraquelli et al. (2005) identify some complementaries among different components, but only slight vertical economies for average-sized firms. Their subadditivity results are in accordance with Gilsdorf (1995), since the results do not support the presence of global subadditivity.

2.1.1.2 Modeling Electricity Distribution Costs

Electricity distribution cost modeling dates back to the 1970s, and most models account only for market variables and system physical characteristics, such as numbers of customers, sales, network length, and load factor (Table 2.2). Detailed urban, geographic, and environmental variables, on the other hand, are overlooked.

12

Author(s) and Year Wells (1977)

Dependent Variable(s) Cost

Neuberg (1977)

Cost

Guldmann (1984a)

Cost (total cost, subcomponents)

Guldmann (1985)


Guldmann (1988)


Nemoto et al. (1993) Clagett (1994)

Variable cost

Salvanes and Tjotta (1994)

Cost

Cost/wage

Independent Variables Customers/network length, distribution plant value per customer, underground lines per network length, isolation index Number of customers, sales, overhead lines, service area, price of labor, price of capital, distributor dummy

Functional form Linear

Study Area and Sample Size British Columbia 54 power districts

CobbDouglas

US (2 data sets) 185 Private and 189 municipal=374 90 Private and 75 municipal=165 cross-section Pacific Gas and Electricity Company 83 communities cross-section Long Island Lighting Company 55 communities cross-section Long Island Lighting Company 110 communities cross-section

Number of customers (residential, commercial-industrial, lighting), sales (residential, commercialindustrial, lighting), density, community residential character Number of customers (residential, commercial-industrial, lighting), sales (residential, commercialindustrial, lighting), density Number of customers (residential, commercial-industrial, lighting), sales (residential, commercialindustrial, lighting), house value, density (residential, commercialindustrial, lighting), units in a structure, housing age Price of labor, price of capital, price of fuel, company dummy

Linear, Log-Log

Number of customers, sales, number of commercial-industrial customers, load factor, wage, network length, employees, value of electric plant, purchases (energy and wholesale power), capital, dummies (distributor, water-electricity-gas customers) Number of customers, wage, labor cost share, electricity cost share, electricity produced, electricity purchased

Translog

Linear, Log-Log

Linear, Log-Log

Translog

Translog

Japan 9 utilities 5-year panel Tennessee Valley 108 municipal and 49 co-operatives 157 total 8-year, pooled

Norway 91 companies cross-section

Continued Table 2.2 Distribution Cost Studies in the Electricity Industry

13

Table 2.2 continued Filippini (1996)

Variable cost

Price of labor, price of energy purchased, capital, load factor, sales

Translog

Filippini (1998)

Cost

Number of customers, sales, load factor, area, price (capital, electricity, labor)

Translog

Salvanes and Tjotta (1998) Yatchew (2000)

Cost

Translog

Filippini and Wild (2001)

Cost

Folloni and Caldera (2001)

Cost

Jamasb et al. (2012)

Total expenditure

Number of customers, wage, labor cost share, electricity cost share, electricity produced, electricity purchased Number of customers, wage, price of capital, load factor, sales per number of customer, network length per number of customer, remaining lifetime of assets public, utility commission Sales, customer density, price of labor, price of capital, high voltage grid dummy, low voltage sales, load factor, average consumption per low voltage customer, land-use (agriculture, forest, unproductive, other ) Number of customers, sales, employed power, network length, length of medium-tension lines, density, employees, network length per employee, network length per customer, employee per customer, territory indicator, amortization per network length, average payroll Network length, network density, energy distribution loss, customer minute lost/number of customers, minimum air temperature, hail, thunder, concrete temperature, willingness to pay, time variable

Cost/customer

14

Switzerland 39 firms 3-year, cross-section Switzerland 39 firms 3-year pooled, cross-section Norway 91 companies cross-section

Translog

Ontario 81 firms 3-year panel

Quadratic

Switzerland 59 firms 380 observations 9-year panel, cross-section

Translog

Italy 147 zones cross-section

Translog

UK 12 networks 8-year panel, cross-section

Customer density is the most commonly used geographic variable. Wells (1977), Huettner and Landon (1978), Guldmann (1984a, 1985, 1988), Filippini and Wild (2001), Folloni and Caldera (2001), Kwoka (2002), and Fraquelli et al. (2005) consider the impact of density on distribution costs, using different measures of density, such as number of customers per network unit length, total population over the service area, etc. However there are few studies that include more sophisticated geographic and environmental variables, such as land use (Guldmann, 1988; Filippini and Wild, 2001; Kwoka, 2002), housing characteristics (Guldmann, 1988), and weather (Jamasb et al. 2012). Wells (1977) is among the first to include geographical and locale-specific variables in an electricity distribution cost model. He uses an isolation index as a measure of distance, terrain and local cost conditions, in addition to a density variable (customers per mile of distribution line). “It is well known that the cost of distribution to customers in low-density rural areas is high relative to more densely settled areas, and, for this reason, subsidies have usually been given in one form or another to provide service to rural areas” (Wells, 1977:88). The density variable appears to have a greater impact on cost than the isolation index, but both of them are statistically significant. The results show that distribution costs at lower densities are almost three times those in higherdensity areas. Filippini and Wild (2001) include land-use, and high and low voltage lines in their average cost model. Agricultural and forest land are statistically significant, with positive coefficients. Jamasb et al. (2012) estimate models of the total costs of distribution companies, with a focus on quality of service. They include weather-related variables (minimum air 15

temperature, number of days with temperature below zero, number of days with hail, and number of days with thunder) in their model. All the weather variables appear to be statistically significant with positive coefficients, which implies that weather has a significant effect on costs. They estimate the cost of quality improvement with a customer-minute loss variable, and observe that different utility configurations affect quality differently. Therefore, improvement policies must consider utility characteristics. There are some studies that estimate distribution cost functions while considering system sub-components. Wells (1977) use the share of underground lines and Neuberg (1977) include overhead lines in their cost functions. Both variables are statistically significant. Guldmann (1984a, 1985,1988) introduces a more detailed analysis by estimating a model for each system sub-component, while considering land-use and housing characteristics. This research can be distinguished from other studies in terms of data used and variables, in addition to disaggregating capital costs and estimating component cost models. The data used are at the community level, either of a single firm or a small number of firms, unlike other utility-level studies (with at least 12 utilities). In Guldmann (1984a), system component costs (conductors, street lights, substations, conduits, poles, and structures and improvements) are modeled using Pacific Gas and Electricity Company data. In Guldmann (1985 and 1988), system components are further disaggregated (overhead and underground lines, and primary and secondary lines) with Long Island Lighting Company data. The latest study highlights the community-oriented character of the distribution system, with the inclusion of median house value, units in housing structure, and housing age. The results indicate that (1) urban characteristics, 16

such as density and residential character, influence the costs of electric distribution facilities, and (2) economies of scale are derived through service densification but not through service expansion. The results also indicate that service by multiple parallel utilities is more expensive than service by a unique utility, pointing to the existence of a natural monopoly. The existence of a natural monopoly is questioned by Henderson (1985), Nelson and Primeaux (1988), Nemoto et al. (1993), Gilsdorf (1995), and Salvanes and Tjotta (1998). Empirical findings confirm that one-firm service is more cost efficient than multifirm production of the output, implying the existence of a natural monopoly, except in Nemoto et al. (1993), where Japanese utilities do not appear to be natural monopolies. The effect of ownership has been the focus of Meyer (1975), Neuberg (1977), and Clagett (1994). These studies compare costs and efficiencies of municipal, cooperative, and private utilities. Meyer (1975) and Neuberg (1977) compare the costs of municipal and investor-owned firms, and their results favor municipal utilities. Claggett (1994) analyzes electricity distribution costs while considering possible multi-utility combinations (with gas and water). His model estimates electricity distribution costs while incorporating the effects of water and/or gas services with dummy variables. The results show that the cost of electricity distribution increases in some utilities when the service is combined with water distribution, which may be explained by the higher installation and maintenance costs of a dual utility. Cooperative utilities appear to incur lower costs than their municipal counterparts. Finally, scale economies are the focus of Weiss (1975), Guldmann (1984a, 1985, 1988), Salvanes and Tjotta (1994), Filippini (1996, 1998), Filippini and Wild (2001), 17

who all observe economies of scale in electricity distribution systems. Filippini (1996, 1998) finds evidence for economies of scale at different output levels, and claims that most firms, particularly medium and small ones, are away from their optimal operation levels. He suggests that small utilities serving adjacent areas should be consolidated, since consolidation, instead of side-by-side competition, could reduce costs and increase efficiency. Nemoto et al. (1993) observe economies of scale in the short run, but diseconomies in the long-run. More recently, Yatchew (2000) provides a more detailed analysis of scale economies, and finds that they vary with the size of the distribution utility, with constant or decreasing returns for larger firms.

2.1.2 Modeling Electricity Distribution Efficiency

The efficiency of electricity distribution has been a focus of research during the past two decades. Different methods have been employed to test efficiency, such as Data Envelopment Analysis (DEA), Stochastic Frontier Analysis (SFA) and Corrected Ordinary Least Squares (COLS). The former is a programming approach, which does not require the a priori formulation of a cost function, whereas the other two are statistical approaches.

2.1.2.1 Electricity Distribution Efficiency Analysis with the DEA Method

DEA, also known as productive efficiency analysis, is the most widely used method in electricity distribution efficiency research. It is a non-parametric linear 18

programming method, which aims to either minimize the inputs with given outputs (input oriented) or maximize the outputs with given inputs (output-oriented) of decision making units (DMUs). The efficiency of DMUs varies from 0 to 1 in the input-oriented case, where the units at the frontier have the score of 1. In the output-oriented case, technical efficiency varies between 1 and potentially infinity. Increasing the number of variables leads to an increase in the number of DMUs at the frontier. Omitting a relevant variable, however, may result in unreliable efficiency results. The number of customers and electricity supplied are common output variables, while other variables can be considered as both inputs and outputs, such as network length (Table 2.3). Comparisons (private firms versus public firms, urban areas versus rural areas, and different countries) have also been carried out using the DEA method. Bagdadioglu et al. (1996) compare public and private firms in Turkey, and conclude that private firms are all efficient, but that public firms’ efficiencies vary. Miliotis (1992) and Chen (2002) observe that utilities serving larger urban areas are more efficient than smaller rural ones. According to Chen (2002), the most important factor affecting efficiency is labor input, but transformer capacity has also a significant impact on efficiency. Hjalmarsson and Veiderpass (1992), on the other hand, show that productivity growth in urban areas is smaller than in rural areas, and provide evidence for economies of density and growth in overall productivity in Swedish electric utilities. Korhonen and Syrjanen (2003) find efficiencies in both urban and rural utilities, but observe diseconomies with high network length and a small amount of energy distributed.

19

Author(s) and Year

Outputs

Inputs

Miliotis

Length of power lines,

(1992)

transformer capacity,

Network length, transformer capacity,

number of customers,

general expenses,

supply, total area

administrative labor,

Method

Study Area and Number of DMUs

DEA

Greece 45

technical labor Hjalmarsson


Network length (high

and

(high and low voltages),

Veiderpass

supply (high and low

and low voltages), transformer capacity,

(1992)

voltages)

employee work hours

Bagdadioglu


Network size,

et al.

supply, service area,


(1996)

maximum demand

labor, network losses,

DEA

Sweden 289

DEA

Turkey 70

general expenses Goto and

Residential sales,

Generation capacity,

Tsutsui

non-residential sales

quantity of fuel used,

(1998)

DEA/AR

Japan and the U.S. (9 Japan, 14 US)

quantity of power purchased, employees

Chen

Number of customers

Network size,

(2002)

(high and low voltages),

length of power lines,

supply (high and low


voltages), total revenue,

general expenses

DEA

Taiwan 22

maximum demand Pacudan and


Number of employees,

de Guzman

electricity sales,

network length,

(2002)

service area

network losses

Korhonen


Operational expenses

and Syrjanen (2003)

supply, quality, network length (different voltage)

Giannikis


Operating expenses,

et al.

network length,

total expenses,

(2005)

supply

security of suppy,

DEA

Philippines 15

DEA

Finland 102

DEA

UK 14

reliability of supply Kinnunen (2005)

Sales/network length

Total cost/total units

DEA

of delivered electricity,

Finland, Norway, Sweden

sales value/total units

(96 ,149,220

of delivered electricity

respectively)

Continued Table 2.3 Efficiency Modeling of Electricity Utilities Using DEA 20

Table 2.3 continued Yu et al.


Operating expenses,

(2009)

network length,

total expenses, duration

supply

of interruptions,

2nd-stage

UK

DEA

12

energy physical loss 2nd-stage non-discretionary variables: Weather Index I: minimum temperature, airfrost, groundfrost, concrete temperature Weather Index II: total rainfall, hail, thunder, maximum temperature, gale

There are some studies that compare efficiencies in different countries or regions. Goto and Tsutsui (1998) focus on the efficiency of the whole industry, instead of on the efficiency of distribution. They incorporate the assurance region (AR) technique into the DEA method in order to eliminate the non-restricted multiplier characteristics of the DEA method. Comparing Japanese and US electricity utilities, they show that Japanese electric utilities are more efficient in technical, allocative and scale terms. The source of overall cost inefficiency in Japanese utilities is allocative inefficiency. Electricity prices tend to be high in Japan because of the excessive amount of capital input, which also leads to allocative inefficiency. Kinnunen (2005) analyzes the economics of electricity distribution systems in Northern Europe. The structural variables turn out to be not explanatory enough to generate a reliable cost model. He suggests that this might be a result from inefficiencies, and therefore decides to run an efficiency analysis. Norway’s efficiency scores are significantly different from those of Sweden and Finland, which is explained by 21

relatively higher wages. Finland appears to be the most efficient country, which may be a result of appropriate regulations. The introduction of environmental variables into DEA models is relatively new. Korhonen and Syrjanen (2003) are among the first do so, and test the possible effects of average snow depth in winter, change in the consumption of electricity, insular areas, and urbanization on utilities’ efficiency. However, they include none of these variables in the final model, which includes energy supply and quality as outputs, and operational expenses as the inputs. Yu et al. (2009) introduce weather parameters into their 2-stage DEA model, in addition to classical input and output variables. They create two indices with weather variables. The first index includes the minimum temperature, airfrost, groundfrost, and concrete temperature (the heat absorbed by concrete during the day, which is an approximation of the urban island effect), while the second index includes total rainfall, hail, thunder, maximum temperature, and gale. In the first stage, DEA is employed to calculate efficiencies using (1) operating expenses, total expenses, duration of interruptions, energy loss, and prices as inputs, and (2) the numbers of customers, the energy delivered, and the network length as outputs. In the second stage, they run Tobit regressions using efficiency scores as dependent variables. The Tobit model is viewed as appropriate by the authors for the second stage since the dependent variable (efficiency score) ranges between 0 and 1. They use weather indices as independent variables to assess their effects on efficiency. The results of some models show that weather indices affect costs and quality in a significant way.

22

There have been discussions regarding the efficiency concept and the limitations of the DEA model. Efficiency can be regarded as an important measure of utilities’ performance. However, cost efficiency may not lead to high service quality. Giannikis et al. (2005) analyze the relationship between the quality of service in electricity distribution firms and cost efficiency. They use DEA and Malmquist indices to test the change in productivity with network improvements, and find that cost efficiency may not result in higher service quality. Mork (1992) discusses the limitations of the DEA method while commenting on Hjalmarsson and Veiderpass (1992). He argues that DEA does not have a behavioral dimension, and may be problematic in decision making, unless productivity is distinguished from scale economies.

2.1.2.2 Electricity Distribution Efficiency Analyses with Parametric COLS and SFA Methods

The Corrected Ordinary Least Squares (COLS) and Stochastic Frontier Analysis (SFA) are parametric modeling methods used to analyze efficiency in electricity distribution utilities. The SFA has two error components: one for the random shocks and the other for the distribution of efficiency. In contrast the efficiency model to which COLS is applied as an estimation strategy is a deterministic – not stochastic – frontier model. COLS estimates parameters by ordinary least squares. “The regression equation is estimated using the OLS technique and then shifted to the efficient frontier by adding the value of the largest negative estimated error to that of estimated intersect (for a cost 23

function).” (Jamasb and Pollitt, 2003:1611). The error term is composed of two parts: one capturing noise and the other capturing inefficiency. The drawback of the method is its assumption of residuals as measuring inefficiency. COLS is employed by Jamasb and Pollitt (2003) and Farsi and Filippini (2004) (Table 2.4). In the former study, the loglinear and translog cost functions are employed. The translog form provides flexibility, and therefore higher efficiency scores, as compared to the DEA method. The latter study shows that model specification is very important, since different models generate different results, and sensitivity analyses are helpful to assess each model’s limitations.

Author(s) and Year

Dependent Variable

Burns and

Operating

Weyman-

cost

Independent Variables Number of customers, sales, maximum demand, area, network length, transformer

Jones (1996)

Functional form

Study Area and Number of Observations

SFA

England,Wales

(Translog) OLS, GLS, MLE

12 utilities 12-year panel

capacity, industrial sales/total, price (capital, labor), density

Filippini

Average

Sales, price (labor, capital),

et al. (2001)

cost

load factor, high voltage network, low voltage sales, average low voltage consumption, density, land-use (agricultural,

SFA

Switzerland

(OLS,

59 utilities

Half normal,

9-year panel

Truncated normal)

forest, unproductive, other) Jamasb

Cost

and Pollitt

Number of customers, sales,

SFA, COLS

network length

(Log-linear,

Countries

translog)

63 utilities

DEA

cross-section

Cost Frontier

Slovenia

(Log-log)

5 utilities

(2003) Filippini et al. (2004)

Cost

Sales, price (labor, capital), density, load factor

6 Eurpean

10-year pooled

Continued Table 2.4 Efficiency Studies on Electricity Utilities Using COLS and SFA 24

Table 2.4 continued Farsi and

Cost/sales


SFA

Switzerland

Filippini

price (labor, capital, power),

(COLS, RE GLS,

59 utilities

(2004)

load factor, area,

RE MLE, FE)

9-year, pooled

SFA

Switzerland

(GLS, MLE,

59 utilities

True RE)

9-year panel

SFA

Switzerland

(Cobb-Douglas)

26 utilities

Pooled

5-year pooled

high voltage network, auxiliary revenues/25% of total revenues dummy, time, forest land/40% of total area Farsi et al.

Cost/sales

(2006)

Number of customers, sales, load factor, price (labor, capital, power), high voltage network, auxiliary revenues/25% of total

Farsi et al.

Cost

(2007)

revenues dummy, area Number of customers, sales, area, price (labor, capital, power), load factor, number of terminal blocks, number of customers/network length, network length

Kopsakangas Savolainen

Cost/sales

Sales, load factor,

Pitt and Lee GLS GLS and Mundlak SFA

price (labor, capital, power)

and Svento (2008)

Finland

(Cobb-Douglas,

76 utilities

Translog)

6-year panel

Fixed Effects and Random Effects Models

SFA, another parametric method to test efficiency, is similar to COLS in terms of cost function specification and efficiency score estimation. SFA is expected to provide higher efficiency scores than COLS due to the inclusion of stochastic error factors. Burns and Weyman-Jones (1996) are among the first to employ SFA in electricity distribution efficiency analysis, and find that the random effects panel data model is 25

superior to the classical cross-section model. They confirm that privatization improves cost efficiencies. Economies of scale are observed in Filippini et al. (2001), and Filippini et al. (2004). The latter study points to inefficiencies in small firms, and recommends consolidation, similar to Filippini (1996, 1998). Filippini et al. (2001), Farsi and Filippini (2004), and Farsi et al. (2006) use the same data set to estimate efficiencies through cost frontier analysis. The former two studies highlight the importance of land-use variables in the cost function. In Filippini et al. (2001), the OLS cost model achieves a fit of more than an 80% fit, and the density, customer structure, and land-use variables are all statistically significant, with the expected signs. The efficiency study shows that the truncated normal model is superior to others. Farsi and Filippini (2004) focus on different models to analyze efficiency and also test model limitations through efficiency score sensitivity analysis. Farsi et al. (2006) suggest that the “true random effect model” can address heterogeneity and efficiency assumptions. All three studies, and a more recent one by Kopsakangas-Savolainen and Svento (2008), point out that those differences in estimations result from unobserved heterogeneity in utility characteristics. Kopsakangas-Savolainen and Svento also suggest that the random effect models generate higher inefficiency scores than the fixed effect ones. Farsi et al. (2007) compare different models using the SFA method, and observe inefficiencies in the Swiss electricity sector no matter which model is used. They show that density and area are statistically significant, and that utilities experience economies of density.

26

2.1.2.3 Electricity Distribution Efficiency: Comparing Parametric and NonParametric Methods

Parametric and non-parametric methods have been compared in some studies (Table 2.5). Hattori et al. (2005) compare electricity distribution efficiencies in the UK and Japan using DEA and SFA. Their detailed analysis covers 3 time periods to measure the effects of time and technical change, and they include environmental factors such as, customer density and load factor in the SFA model. The results demonstrate that UK firms are more efficient than the Japanese ones in general, with some efficiency variations in different time periods. The study also shows that applying different methods (DEA and SFA) may result in different outcomes. For instance, according to DEA results, efficiency change and technical change are significantly different in the two countries. According to SFA results, all differences are attributed to efficiency change since the results do not capture any difference in technical change.

Author(s) and Year

Outputs

Inputs

Method


Jamasb


Operating expenses,

DEA

6 Eurpean

and

network length,

total expenses,

COLS

Countries

Pollitt

supply

network length,

SFA

63 observations

(2003)

transmission and distribution losses

Hattori


Total expenditure (for

DEA

UK and Japan

et al.

network length,

individual efficiency)

SFA

(12 UK, 9 Japan)

(2005)

electricity sales

operating expenditure

maximum demand

(for comparison)

Table 2.5 Efficiency Studies of Electricity Utilities: Comparing Different Methods 27

Jamasb and Pollitt (2003) compare benchmarking techniques and method choice in efficiency assessment. They show that efficiencies vary with different methods and model specifications, and that each model has advantages and disadvantages. The selection of appropriate input and output variables is crucial in DEA models, while model specification is very important in statistical approaches.

2.1.2.4 Electricity Distribution Efficiency Studies Using Profit Function

Profit, instead of cost, has also been used to assess efficiency. Claggett et al. (1995) use a profit function to test efficiency in the Tennessee Valley Area. They estimate profit functions for 74 municipal and 45 cooperative electricity distributors, and test whether municipal utilities and cooperatives are as efficient as private firms. Profit is defined as a function of prices of inputs, labor, purchased electricity, capital (length of distribution lines), percent line loss, service area density, and presence of municipal or cooperative distributors. The results show that municipal utilities and cooperatives aim to minimize their costs while none of them maximizes their profits.

2.2 Natural Gas Economics

The issues related to natural gas distribution are similar to those related to electricity distribution. Natural gas cost modeling and efficiency assessment studies also exist, although not as extensively. Cost research starts with production, then shifts to

28

transmission and distribution. The efficiency of natural gas utilities has been considered in only a very few studies.

2.2.1 Natural Gas Distribution Cost Models

Research on the economics of the natural gas industry dates back to the 1940s. However, the initial studies focused on gas production economies (Verhulst, 1948; Gribbin, 1953; Durer and Slater, 1977). Modeling distribution costs, on the other hand, attracted the attention of researchers much later on. A summary of research on natural gas distribution costs is presented in Table 2.6. Distribution costs have been defined as a function of input price and output variables, as well as spatial factors. However, some researchers have not considered spatial/environmental variables. Klein (1993) estimates cost functions with only input prices and outputs in different customer classes: residential, commercial, and industrial. Most researchers, however, have realized the relevance of environmental and locale-specific factors, and have integrated them into their models. “Varying physical conditions of the supply area (or network configurations), let alone output quantity, would greatly affect the cost behavior. Therefore, it does not make sense to estimate the cost function with only supply and quantity and input prices.” (Kim and Lee, 1995, 247-248).

29

Author(s) and Year

Dependent Variable(s)

Guldmann

Cost

Sales (residential, non-residential)

(1983)

(capital)

number of customers (residential,

Independent Variables

Functional form

Study Area and Sample Size

Log-Log

Long Island, East Ohio, 101

non-residential), population density Guldmann

Cost

Sales (residential, non-residential)

(1984)

(capital)


cross section Log-Log

non-residential), population density Guldmann

Cost

Sales (residenial, non-residential)

(1985)

(capital)


cross section Box-Cox

California, 240

population density, load factor Cost

Residential sales (heating,

(1989)

(capital)

non-heating), commercial sales,

cross section Box-Cox

Cost

(1993)

Total operating costs,

Kansas,Nebraska, Iowa, Minnesota

industrial sales, population density Klein

Long Island, Ohio, Iowa,

non-residential), wage rate Guldmann

Long Island, East Ohio, 101

65 Translog

Tennessee

sales (residential, commercial,

6

industrial, other), purchased gas

13-years panel

expense per mcf sold, average wage of utility employees, sales in peak month Kim

Cost

Customer density, average

and Lee

customer size, supply rate, prices

(1995)

of input factors, actual output

Fabbri

Cost


et al.

sales/customers, network length,

(2000)

network length/customers,

Translog

Korea 7 5-years panel

Translog

Italy 62 2-years panel

concentricity ratio of population, average altitude, employees, customers/employee, labor cost/employee, external cost/network length, value of equipment/network length, distribution costs, shares of external costs, labor costs, depreciation in distribution costs Bernard et al.

Cost

Maximum daily demand, pipeline length, regional dummy

(2002)

Box-cox

Quebec

Log-linear

131 cross section

Table 2.6 Distribution Cost Studies in the Gas Industry 30

Guldmann (1983, 1984, 1985a, and 1989) is one of the first to analyze gas distribution costs while considering the multiproduct, multidimensional character of the system. He suggests that empirical econometric modeling provides for a better understanding of the gas industry. In earlier studies, Guldmann (1983, 1984) defines cost as a function of the number of customers and sales in the residential and non-residential sectors as well as of population density. He observes the positive effects of residential sales and numbers of residential and non-residential customers on costs, while density has a negative effect. He shows that service densification contributes to economies of scale in a significant way, while increasing market size contributes only slightly to economies of scale. Guldmann (1985a) focuses on the capital component of the total cost, and models the capital costs of the distribution system as a function of residential and non-residential sales, numbers of residential and non-residential customers, population density, load factor, and wage. Capital needs increase with increasing sales, numbers of customers, and wage rate, but decrease with increasing density and load factor. In accordance with the earlier studies, there is evidence of economies of density as a result of service densification. Cost savings through densification are greater than through expansion, because densification requires comparably fewer lines and enables a more efficient use of existing lines. Guldmann (1989) models capital input as a function of sales in different sectors and density. All sales variables have positive signs, but the density coefficient appears to be zero, which is explained by limited density variations in the sample area. The results show that marginal capacity costs are very high for residential-non-heating customers as a result of the fixed costs per customer that have to be spread over low consumption levels. Efficiencies appear to vary across different 31

sectors, with larger communities tending to be subsidized by smaller ones. Large industrial and residential customers benefit from the cost structure whereas commercial and small industrial customers are at a disadvantage. Kim and Lee (1995) include customer density and average customer size as variables in their cost model, and all of the variables – density, customer size and supply rate – are negatively related to costs. They suggest that lower values of these variables characterize a better network structure, and a better network leads to lower costs. Fabbri et al. (2000) include the average altitude of the study area and the level of urbanization, in addition to density. Cost is defined as a function of the amount of gas delivered, numbers of customers, territorial variables, input prices, a time shift variable, and a shift variable for private firms. The number of customers appears to be more significant than the volume of gas delivered. Input prices are all significant, with expected signs. Unlike in most other studies, they find that population concentration has a positive coefficient, which is explained by the diseconomies resulting from urban congestion. Altitude has a positive sign, which implies higher costs in mountains. They observe that economies of scale vary with different output levels, number of customers, density, concentricity ratio (as a measure of the population concentration in inhabited areas), average altitude, and input prices. However they conclude for constant returns to scale because the minimum and maximum values of scale economies are very close to 1. Bernard et al. (2002) use maximum daily demands instead of sales as explanatory variables. They compare cost differences in different regions in Canada. Quebec, the largest and the oldest region appears to have the highest costs, which is explained by area-specific pavement requirements. 32

2.2.2 Natural Gas Distribution Efficiency Models

Recent research has focused on efficiency in the natural gas industry rather than cost function modeling. DEA, SFA or profit functions have been used to analyze the efficiency of gas utilities, in a way similar to electricity distribution. Different approaches may generate different results in efficiency scores, and it is therefore important to account for the advantages and limitations of each method.

2.2.2.1 Natural Gas Distribution Efficiency Analysis with the DEA Method

There are very few natural gas efficiency studies. Hollas et al. (2002), Erbetta and Rappuoli (2008), and Erturk and Turut-Asik (2011) use DEA to analyze efficiency in natural gas distribution systems (Table 2.7). Hollas et al. (2002) include gas supplied to different customer groups (residential, commercial and industrial), but no environmental variable. The eras before and after market regulation are compared to test for the impact of competition on efficiency. The findings indicate that increasing competition results in the loss of economies of scale, but this does not affect the overall economic efficiency and the overall technical efficiency in a negative way. Erbetta and Rappuoli (2008), on the other hand, include customer density in their model. The sample is divided into two subsets in terms of density, as higher densities are expected to lead to more efficiency. The results show that technical efficiencies of the firms are equal on the average, which can be a result of a lack of competition in the distribution system. Scale efficiencies of the firms are also equal on the average, but with higher values than technical efficiency. 33

Customer density appears to have a direct effect on the results, as expected, suggesting that merging smaller units would increase productivity.

Author(s) and Year

Outputs

Inputs

Hollas et al.

Gas delivered (residential,

Capital, labor,

(2002)

commercial, industrial)

purchased gas

Erbetta and


Capital, labor, materials,

Rappuoli

gas delivered,

maintenance,

(2008)

customer density

other services

Erturk and

Consumption (total,

Network length, length

Turut-Asik

residential, industrial),

of pipeline (polyethylene,

(2011)

number of customers,

steel), employees

peak demand

total cost, operating

Method


DEA

U.S. 33

DEA

Italy 46

DEA

Turkey 38

cost, capital cost Environmental factors: Average winter temperature, inverse apartments/building ratio

Table 2.7 Efficiency Studies on Natural Gas Utilities Using Non- Parametric Methods (DEA)

Erturk and Turut-Asik (2011) consider more detailed environmental factors in their efficiency model, such as winter temperature and building stock characteristics. These environmental factors have significant effects on the efficiency scores of DMUs, particularly those operating in environmentally unfavorable areas. They suggest that the inclusion of environmental factors into a DEA model to leads to more reliable efficiency scores. They also analyze the effect of the socio-economic development levels of the regions, and find that developed areas are more efficient. The comparison of public and 34

private firms shows that public firms are more efficient, and the comparison of small and large firms shows that small firms are inefficient.

2.2.2.2 Natural Gas Distribution Efficiency Analysis with the Parametric SFA Method

Rossi (2001) and Farsi et al. (2007) employ the stochastic frontier analysis (SFA) to assess efficiency (Table 2.8). In both studies, different functional forms are compared and the consistencies of the estimates are discussed across different models. Rossi uses the number of customers as the dependent variable, and examines the effect of privatization on technical efficiency, concluding that technical efficiency improves with privatization. Farsi et al. estimate total costs and include customer density and area size as locale-specific variables. Customer density has a negative effect, whereas area size has a positive effect on costs. They employed the GLS-Mundlak model to test for economies of scale and density. The results confirm the presence of economies of density and weak or insignificant scale economies.

35

Author(s) and Year


Rossi

Number of

(2001)

customers

Farsi et al. (2007)

Cost

Functional form


Network length, sales (total,

SFA

Argentina

residential/total), area, time,

(OLS)

40

number of employees,

(COLS)

5-years panel

maximum demand

(MLE)


Energy value of delivered gas,

SFA

Switzerland

prices (capital, labor, energy),

(Pooled MLE)

129

load factor, customer density

(Pitt and Lee MLE)

5-years panel

number of terminal blocks, area size, network length

(GLS) (GLS and Mundlak)

Table 2.8 Efficiency Studies on Natural Gas Utilities Using Parametric Methods (SFA)

2.2.2.3 Natural Gas Distribution Efficiency Analysis with Profit and Cost Functions

Asides from parametric and non-parametric techniques, profit functions have also been used to assess efficiency in natural gas distribution from a pricing point of view. Hollas and Stansell (1994) compare municipal and private utilities. They define profit as a function of input prices, fixed capital input and the ownership, and find evidence that municipal firms are more price-efficient than private firms, while private firms are more technically efficient than municipal ones. Overall, they conclude that private firms are more efficient as a result of higher levels of technical efficiency. Hollas (1999) analyzes the price structure in the residential, commercial and industrial sectors, and examines the effect of market structure. He shows that deregulation has resulted in decreasing prices for the industrial sector when compared to the other two sectors, while the residential sector appears to be the most disadvantaged one. 36

Casarin (2007) uses a different approach to assess the cost efficiency of natural gas systems in Argentina and Britain. He analyzes industry configuration and breaks transmission and distribution industries into sub-components. The study questions whether it is more efficient for a utility to operate as a whole system or as subcomponents in the presence of other firms. The cost function, however, does not estimate exclusively the distribution costs, but the distribution and transmission costs together. Five scenarios are created to test efficiency: transmission, distribution and supply integrated; transmission and distribution integrated; distribution and supply integrated; transmission only, and supply only. The transmission and distribution output is the volume of natural gas, while the supply output is the number of customers. The study suggests that firms are more efficient when they have separate downstream components.

2.3 Economics of Other Utilities

Network utilities are not limited to electricity and natural gas. Research on other utilities, such as water and telephone, can provide insight into the economics of electricity and natural gas distribution (Table 2.9). Kim (1987) considers the multi-product character of the water industry, but neglects urban, geographic and environmental factors. The study provides evidence of overall constant returns to scale. With a focus on product-specific scale economies, the non-residential sector enjoys substantial economies of scale. The residential sector, on the other hand, is characterized by diseconomies of scale. The marginal cost of nonresidential water usage appears to be smaller than the residential one. Bhattacharyya et al. 37

(1995) analyze the efficiency of water networks. They define a shadow cost function with hedonic specification. The empirical findings point to overutilization of energy as compared to labor, with labor and energy being substitutes for each other. Fabbri and Fraquelli (2000) address water network costs and use four hedonic variables: number of customers, density, cost of water input purchased, and treatment costs. The hedonic variables are all statistically significant. They claim that model specification and functional form modify the results of the economies of scale analysis. Cubukcu and Guldmann (2008) point to the effects of urban and geographic variables on telephone network costs. Locale-specific geographical variables include the area of the service territory, land uses, street patterns, population density, soil types, slope, and the spatial partitioning of a company into local exchanges. The authors develop an index defining land use prominence in the service area. The results confirm the importance of geographical factors. Economies of density appear to be exhausted at a very high output level, which is beyond the maximum output in the sample, but there is one company having diseconomies of scale due to its service area geographic conditions. Spatial expansion leads to exhaustion of economies of scales at lower output levels than those for economies of density exhaustion. Six companies appear to have diseconomies of scale in the sample. The results point to the existence of a natural monopoly.

38

Author(s) and Year


Kim (1987)

Water cost


Functional form


Translog

US

Sales (residential, nonresidential), price (labor, capital, energy),

60 utilities

capacity utilization,

cross-section

network length, labor share, capital share, energy share Fabbri and

Water cost

Number of consumers,

Translog

Italy

Fraquelli

water purchase/total cost,

(hedonic,

173 utilities

(2000)

treatment/total cost, density,

non-hedonic)

cross-section

price (labor, capital, energy)

Cobb-Douglas (hedonic, non-hedonic)

Bhattacharyya

Water

Price (energy, labor), water

et al. (1995)

shadow

input, capital, density,

cost

service diversity dummy,

Non-linear

Nevada 26 utilities cross-section

ownership dummy, treatment dummy, metered connections, connections/network length Cubukcu and

Telephone

Labor cost, telephone units,

Guldmann

cost

price (labor, capital),

(2008)

technology, area, density, house rent, built-up dominance

Translog

New York State 41 utilities cross-section

index, street segment, inter-exchage link, soil, slope

Table 2.9 Cost Modeling for Water and Telephone Utilities

2.4 Economics of Multi-Sector Utilities

There are utilities that operate as multi-sector firms, providing different combinations of electricity, natural gas, and water services. These utilities are expected to have an incentive to operate in such a way. Researchers have examined the presence of 39

economies of scope, that is, the benefits resulting from the joint provision of these multiple urban services (Table 2.10). Economies of scope are expected to stem from use of common infrastructure and services, as well as operating in the same geography.

Author(s) and Year


Mayo

Cost

(1984) Chapell

Cost

Independent Variables Electricity output,

US 200 utilities cross section

Electricity output using fossil fuel,

(1986)

of fuels, gas output

(1987)

Quadratic

dummies (electricity, gas) electricity output using all types

Cost


gas output,

et al.

Sing

Functional form

Quadratic

US 88 utilities cross-section

Electricity output, gas output, prices (labor,

Generalized

US

translog

108 utilities

capital, fuel), density, cost

cross-section

shares (labor, fuel, capital) Fraquelli

Cost

Electricity output,

et al.

gas output, water output,

(2004)

prices (labor, other), cost shares (labor, other)

Translog,

Italy

Generalized

90 utilities

translog,

3-year pooled

Seperable quadratic, Composite

Piacenza

Cost

Electricity output,

and


Vannoni

prices (labor, other)

(2004)

Translog,

Italy

Generalized

90 utilities

translog,

3-year pooled

Seperable quadratic, Composite

Farsi

Cost

Electricity output,

Quadratic

et al.


(2008)

prices (labor, capital, electricity, gas), density, dummies (electricity, gas, water)

Table 2.10 Multiple-sector Utilities Cost Modeling 40

Switzerland 87 utilities 9 year panel

According to Mayo (1984), economies of scale provide implications about the cost structures of single-sector firms, such as electricity only firms or natural gas only firms. However, in case of a multi-product provision, such as firms serving both electricity and natural gas, economies of scope should be considered. Mayo uses a quadratic cost function to analyze the costs of various service combinations: electricity only, natural gas only, and electricity and natural gas combined. The results point to economies of scope at lower output levels, and diseconomies of scope at higher output levels. Diseconomies of scope are claimed to result from inefficiencies due to the absence of competition. The interaction variable has a positive sign, which indicates that interproduct complementarities are not available in these regulated firms. Chapell et al. (1986) follow Mayo (1984), and take his study one step further by adding a technology variable. Technology is defined for electricity firms using fossil fuels and nuclear power. The results are similar to those of Mayo (1984), in terms of the presence of regions reflecting economies and diseconomies of scope. Diseconomies of scope and scale decrease for firms using fossil fuels. Chapell et al. conclude that, economies of scale and scope can be related to the technology used, as well as to the size of the firm. Fraquelli et al. (2004), and Piacenza and Vannoni (2004) use the same data to analyze scale and scope economies in 90 utilities with various combinations of gas, water and electricity service. The sizes of the utilities vary. Cost is considered as a function of the quantities of gas, electricity, and water, and the prices of labor and other inputs. The comparison of four models (standard translog, generalized translog, quadratic and composite) show that the composite cost model, which considers output and price 41

interactions, is the most suitable model in both Fraquelli et al. (2004) and Piacenza and Vannoni (2004). The highest cost advantage appears to be gained in gas-water combinations in small firms. Small to medium firms are enjoying global and productspecific economies of scale and scope, while larger firms are experiencing aggregate economies of scale and scope. The authors suggest that small firms engage in multiple utilities to reduce their costs. Farsi et al. (2008) analyze the economies of scale and scope of 87 utilities serving various combinations of gas, electricity, and water. They utilize the quadratic cost functions estimated with the GLS and random-coefficient (RC) methods. Customer density is introduced into the cost model, in addition to electricity, gas and water outputs, and input prices. The density variable is statistically significant, with the expected negative sign. These empirical results provide evidence for both scope and scale economies in the majority of the multi-utilities. As in previous studies, utilities that benefit from economies of scope savings are the small ones. The authors interpret the results as proving the existence of a natural monopoly structure in multi-utilities. Sing’s (1987) findings are contradicting the others, showing diseconomies of scope at the mean combination level, although there are some regions with economies of scope. Sing believes that other incentives must be present for utilities experiencing diseconomies of scope, to enable them to continue their multi-product service provision.

42

2.5 Concluding Remarks

Research on urban energy distribution systems has developed throughout the years, focusing on costs, scale and scope economies, and efficiency issues. These sectors tend to reflect natural monopoly characteristics, and economies can be achieved primarily through densification. There are some contradictory studies that favor competition in the downstream component of the industries. The literature on electricity and natural gas distribution costs often lack localespecific characteristics. The only variables that have been taken into account are service area and density in most research. Land-use, housing characteristics, urban structure, topography, soil types or weather variables, on the other hand, have been considered only in a limited number of studies. The literature on efficiency is relatively new. Parametric and non-parametric techniques have been employed to assess the efficiency of energy utilities. Efficiency scores are highly dependent on the selected input and output variables. In non-parametric methods, the selection of the functional form also plays an important role. Economies of scope analyses are ambivalent. Some studies point to the presence of cost benefits from joint service provision in small multi-sector utilities, whereas some claim the presence of diseconomies of scope. Environmental and geographic variables are widely overlooked in these studies, while the costs are defined only as a function of outputs.

43

CHAPTER 3 METHODOLOGY

Neoclassical economic theory provides a basis for selection of the variables and modeling. Aggregate and disaggregate capital investment costs are modeled in electricity and natural gas distribution systems. A multi-utility cost model is also provided to analyze the economies of scope.

3.1 Introduction

The economics of electricity and natural gas distribution deal with cost modeling, scale economies and monopolistic structure, efficiency measures, and economics of multi-utilities. There are two literature streams of research on the economics and planning of energy distribution systems: engineering/optimization models and econometric models. In the electricity industry, for instance, engineering and optimization models attempt to find the sizing and least-cost siting of substations, and are useful to design an actual system, but they cannot, by their very nature, reveal industrywide properties. Econometric models, in contrast, can capture multi-product and multidimensional character of the output, and stem from traditional economic theory. 44

Hence, they are more appropriate to analyze the cost structure of these distribution systems.

3.2 Cost Modeling in Neo-classical Economic Theory

A firm is assumed to produce outputs with a given set of inputs, such as labor (L), capital (C), and energy (E). The production function, , is represented in Equation (3.1)

(3.1)

The production function can be extended to multi-product and multi-dimensional industries, using the transformation function (Eq. 3.2). Electricity and natural gas are considered such industries, due to their serving different customer groups: residential, commercial, industrial, lighting (in electricity), etc. There may be even further variation within a customer group, such as small and large industries. Let Q represent the vector of these outputs. Feasible substitutions of inputs and outputs are assumed to be summarized by the transformation function.

(3.2)

The vector of site-specific variables, SH, can be integrated to the transformation function:

(3.3) 45

Public utilities serve clearly delineated territories at regulated prices, and therefore try to minimize their costs:

min pK K + pL L + pE E subject to

=0

(3.4)

As the price of labor, PL, the price of capital, PK, and the price of energy, PE, are given, a public utility firm searches for L, K, and E that minimize C subject to constraint (Eq. 3.2) and a given output vector Q. Indeed, electricity and natural gas companies are not monopsonists. There are many employers in the region, and the labor market must be viewed as competitive. Thus, firms are price-takers in the labor market, i.e., view the wage rate as given. The price of energy (electricity or natural gas) can also be viewed as given. In 1980, this price was heavily regulated, and electricity/natural gas companies had no freedom in price fixing. This is no longer the case today for the commodity itself, but it is still true for transportation (transmission and distribution) costs. The price of labor is the average wage rate (wage per employee). The price of capital is not available, but it is not expected to have much variation within the small geographic area of this study. The general form of the minimized cost function is:

pK K* + pL L* + pE E*

(3.5)

where P = (pK + pL + pE) and K*, L*, E* = optimal input values. 46

However, the focus of this research is the modeling of the capital costs of the distribution system. The capital cost function is then:

(3.6)

The primary contribution of this research is the inclusion in the vector Q of various site-specific factors, in particular (1) urban form and development captured by density, built-up area, street pattern, and housing stock; (2) local physical geography, captured by soil and topography; and (3) socio-economic structure, captured by income and house value. These variables are often termed hedonic variables. There are various functional forms used in total cost and capital cost models in the literature. Both total cost functions and capital cost functions are presented in the following part, although this research focuses only on capital costs.

3.2.1 Total Cost Function

The choice of the functional form plays an important role in the economic analysis of distribution utilities. Kwoka (1996) indicates that the two most used functional forms for total cost function estimation are the translog and quadratic forms. The translog functional form have been used by Nemoto et al. (1993), Clagett (1994), Salvanes and Tjotta (1994, 1998), Filippini (1996, 1998), Yatchew (2000), Folloni and Caldera (2001), and Jamasb et al (2012) for electricity; Klein (1993), Kim and Lee (1995), and Fabbri et al. (2000) for natural gas; Kim (1987), and Fabbri and Fraquelli 47

(2000) for water; Cubukcu and Guldmann (2008) for telephone; and Sing (1987), Fraquelli et al. (2004), and Piacenza and Vannoni (2004) for multiple utility cost modeling. The use of the quadratic form is generally preferred in multiple utility studies: Mayo (1984), Chapell et al. (1986), Piacenza and Vannoni (2004), and Farsi et al. (2008). These functional forms are appropriate for the multi-product character of total costs. The general form of a translog cost is presented in equation (3.7):

[ [

][

]

[

]

[

] ][

]

[ [

][

] ]

(3.7)

3.2.2 Capital Cost Function

Capital cost functions for electricity and natural gas distribution are estimated in this study. Capital cost includes the physical plant component of the distribution system, such as conductors, conduits, poles, transformers, and services for electricity, and mains, measurement stations, and services for natural gas. Linear, log-linear, log-log, and Box-Cox functional forms have been used in the literature on capital cost function estimations, such as Guldmann (1983, 1985). The general expression of the capital cost function (3.6) can be expanded by explicitly adding the site-specific hedonic variables

.

(3.8) 48

The expanded general equation (3.8) can be estimated with (1) linear, (2) log-log, and (3) Box-Cox functional forms:

∑

(1)

∑

∑

(2)

∑

∑

(3)

∑

∑

∑

∑

∑

(3.9)

∑

(3.10)

∑

(3.11)

where the Box-Cox (1964) transformation is: {

(3.12)

Functional forms are assumed to have normally distributed error terms (not explicit in Eqs. 3.9 – 3.11), with zero means and constant variances, and not conditional on any independent variable. The regression results obtained while using these functional forms can be compared, using log-likelihood values. With

in the linear (additive) form, or

in the log-log form, The log-likelihood test follows a Chi-square distribution (

with two degrees of freedom at the (

[

level of significance:

]

(3.13) 49

In this research, the log-log or Box-Cox functional forms are selected and compared for various network components, such as conductors, conduits, transformers, and poles for electricity, and mains for natural gas. The additive form is preferred when the component is directly related to a customer, and does not have a system wide function, such as services for both electricity and natural gas, and street lighting for electricity. In such cases, costs are directly related to the number of customers in the relevant sectors, and they are separable across the various sectors (residential, commercial-industrial, lighting).

3.2.3 Elasticity

The elasticity measures the percent change in the dependent variable resulting from a one percent change in one of the independent variables, keeping the other independent variables constant. The general form of the point elasticity of the dependent variable (y) with respect to (x) is presented in Equation (3.14):

(3.14)

The point elasticities of the (1) linear (additive), (2) log-log, and (3) Box-Cox models are presented in Eqs. 3.15-3.17, where the coefficient of the independent variable, x, is assumed to be .

50

(1)

(3.15)

(2)

(3.16)

(3)

(3.17)

where is the Box-Cox parameter of the independent variables, is the Box-Cox parameter of the dependent variable.

The elasticity in the log-log model is constant, while the elasticities of the two other forms vary with different values of y and x.

3.2.4 Economies of Scale and Density

The value of the point elasticity is considered as the measure of economies of scale in a firm with a single output. In the case of multiple-output firms, ray economies of scale are measured by the sum of the elasticities of all outputs:

∑

(3.18)

51

where

implies economies of scale,

implies diseconomies of scale, and

implies constant returns to scale. In this research, ray economies of scale can be interpreted as economies under service expansion with a constant density. Economies of density differ from economies of scale by considering both service expansion and densification at the same time. When service expands within a fixed territory, there is densification of the market. The measure of economies of density is:

(3.19) where

is the elasticity of density.

3.2.5 Economies of Scope

Economies of scope are observed in a firm when the joint production of two or more products leads to a smaller total cost than the sum of the costs of producing each product by a separate firm. Economies of scope in case of two products y1 and y2, exist if:

∀

(

(3.20)

The degree of economies of scope may be measured by the following ratio:

[

]

(3.21)

52

Positive values of

imply economies of scope, whereas negative ones imply

diseconomies of scope. “For an industry that enjoys no economies of scope, a multiproduct firm can be broken up into several specialized firms without any increase in cost and, perhaps, even with some savings” (Baumol et al., 1982: 73).

53

CHAPTER 4 DATA SOURCES

This chapter introduces and explains the data used in this research. There are four general categories of data used: company data, census of population, economic censuses, and geographic data. Company data includes sales, number of customers, and investment information on four electricity and natural gas provider companies in the State of New York. Company data are geographically limited to the territories served by these companies. Census of population and geographic data are available for the whole State of New York. Economic Census data are available only in jurisdictions with population greater than 2,500.

4.1 Company Data

Company data characterize four electricity and gas utilities serving in the State of New York: (1) Central Hudson Gas and Electricity Company (CH), (2) Long Island Lighting Company (LILCO), (3) Niagara Mohawk Power Corporation (NM), and (4) Orange and Rockland Utilities (OR). These data include investment plant costs and numbers of customers and sales over the period 1977-1983. A cross-section of these data 54

for 1980 is used in this research to match census data. There were other electricity and gas companies operating in the State of New York, in areas where these four companies did not operate (e.g. Consolidated Edison of New York in New York City and Westchester County, and National Fuel Gas Distribution Company in Buffalo and Western New York State).

4.1.1 Plant Investment Data

Plant investment data include 1980 replacement and historical costs for different plant components. Plant data is available only for the tax districts covered by the four above-mentioned four companies. The New York State Board of Equalization and Assessment (NYSBEA) groups plant costs in six major categories: (1) intangible plant, (2) production plant, (3) storage plant, (4) transmission plant, (5) distribution plant, and (6) general plant. These categories have further sub-categories. This research focuses on distribution costs; therefore only the distribution plant and its sub-categories are considered in this analysis. In the electricity sector, 1014 tax districts have distribution plant data (Figure 4.1). The area with electric plant information covers around 64% of the whole State of New York, while the total population of these tax districts represents almost 43.3% of the State’s population. Niagara Mohawk serves 84% of the area and 51.7% of the population covered by the four companies. It also incurs the largest distribution costs (Table 4.1). LILCO, although covering a much smaller share of the area (3.5%), incurs the second highest cost, because of its large population (35.3%). 55

Figure 4.1 Geographic Distribution of the Tax Districts of the Selected Electricity Companies

Company Central Hudson

Historical Electricity Distribution Plant ($) Number of Districts Minimum Maximum Total Mean Std. Deviation 85

92 10,548,778

134,079,177 1,577,402

1,866,997

LILCO

119

849 74,846,657

540,044,378 4,538,188

11,408,908

Niagara Mohawk

753

106 82,548,000

842,706,684 1,119,132

3,838,663

57

1,987 113,268,708

113,268,708 1,987,170

3,342,909

Orange and Rockland Total

1,014

1,630,098,947

Table 4.1 Descriptive Statistics for the Costs of the Historical Electricity Distribution Plant ($)

The total distribution plant costs vary accross tax districts (Figure 4.2). In most tax districts, this cost is less than $2.7 Million. However, it may exceed $45 Million, 56

particularly in populated areas. Costs are the greatest in Hempstead, Brookhaven, and Buffalo, which are the most populated three areas within the State, after New York City.

Figure 4.2 Geographic Distribution of Electricity Distribution Total Costs ($1,000)

Electric plant distribution costs data include information on 10 sub-categories (Table 4.2). These 10 accounts have each sub-categories. The code for each item has eight digits. The first digit indicates type of the utility: 1 for electricity, and 2 for natural gas. The next four digits indicate the account sub-category. For instance the 5-digit code 10361 indicates “Structures and improvements” of electricity plant (the first digit 1 indicates electricity plant). The sub-groups are coded using additional 3 digits. For 57

instance “Structure and improvements” includes 2 groups: “Structure and improvements below 69KV”, with the code 10361010, and “Structure and improvements 69KV and up”, with the code 10361020.

Account 10361 10362 10364 10365 10366 10367 10368 10369 10371 10373

Percentage of Total Cost 1.7 13.4 17.8 19.8 5.3 8.6 24.8 2.6 0.1 5.8

Description Structures and improvements Station equipments Poles, towers &fixtures Overhead conductors & devices Underground conduit Underground conductors and devices Line transformers Services Installations on customers premises Street lighting-signal

Table 4.2 Electricity Distribution Plant Accounts and Their Shares of the Total Cost

The line transformers group, which includes transformer stations, overhead line transformers, and underground line transformers, represents the largest share (24.75%), followed by the overhead conductors and devices (19.78%), and poles, towers and fixtures (17.83%). In the natural gas sector, 436 tax districts have distribution plant data (Figure 4.3). The area with natural gas plant data covers 18.3% of the whole State, while the total population of these tax districts represents almost 29.6% of the State’s population.

58

Figure 4.3 Geographic Distribution of the Tax Districts of the Selected Natural Gas Provider Companies

Total plant historical costs display variations across companies (Table 4.3). Niagara Mohawk covers 70.4% of the total area covered by the 4 companies with the largest number of tax districts (243). LILCO has also large plant costs due to its high population concentration. The tax districts with the largest natural gas distribution costs differ from the electricity highest-cost districts, except Hempstead (Figure 4.4). Syracuse and Islip are among the highest natural gas investment tax districts. The total natural gas investment costs are smaller than the electricity ones, which can be explained by consumer choices. 59

Electricity is a necessity for every area, whereas natural gas preference depends on consumer behavior and market conditions.

Company Central Hudson

Number of Historical Natural Gas Distribution Plant ($) Districts Minimum Maximum Total Mean Std. Deviation 37

230

LILCO

113

Niagara Mohawk

243 43

727

Orange and Rockland Total

3,507,261

678,255

827,527

80

27,816,426 157,057,359 1,389,888

3,763,891

91

29,811,833 276,347,744 1,137,234

2,669,429

11,976,390

2,368,122

436

25,095,444

53,059,652 1,233,945 511,560,199

Table 4.3 Descriptive Statistics for the Historical Natural Gas Distribution Plant ($)

Figure 4.4 Distribution of Natural Gas Distribution Plant Total Costs ($1,000) 60

The natural gas distribution plant includes different accounts, which are further subdivided into sub-categories (Table 4.4). The coding is similar to that for electricity, the main difference being the first digit (2 for natural gas). The next four digits indicate account category, and the additional 3 digits refer to sub-groups. The code 20376 stands for “distribution mains”, which include “steel mains” (20376100), “plastic mains” (20376200), “iron mains” (20376300), etc. The “Distribution mains” group makes up 78% of the distribution plant.

Account 20375 20376 20377 20378 20380 20385

Description Structures and improvements Distribution Mains Compressor station equipments Measurement and regulation station equipments in service Services Industrial measurement and regulation station equipment

Percentage of Total Cost 0.5 77.7 0.1 2.9 18.7 0.1

Table 4.4 Natural Gas Distribution Plant Accounts and Their Shares of the Total Cost

There are 434 tax districts which have both electricity and natural gas plants, 2 tax districts with only natural gas plant and 580 districts with only electricity plant. Among the tax districts with both plants, 37 are served by Central Hudson, 113 by LILCO, 241 by Niagara Mohawk, and 43 by Orange and Rockland.

61

4.1.2 Customers and Sales Data

Customers and sales data are available for the four companies. Data include revenues, sales, and number of customers in different sectors: residential, commercialindustrial in electricity and natural gas, as well as lighting in electricity. However, as for plant investments, these data are also limited to a number of tax districts. In the electricity sector, 257 tax districts have customers and sales data (Figure 4.5), whereas in the natural gas sector, 210 tax districts have customers and sales data (Figure 4.6).

Figure 4.5 Geographic Distribution of Tax Districts with Electricity Sales and Customers Data

62

Figure 4.6 Geographic Distribution of Tax Districts with Natural Gas Sales and Customers Data

The residential sector has the largest number of customers, whereas the commercial-industrial sector has the largest sales (Table 4.5). The commercial and industrial sectors are merged together in company data, except for Central Hudson. Therefore, the commercial and industrial sectors of Central Hudson are aggregated in order to match the data of the other companies. Electricity is used in space heating, cooking and in household appliances in the residential sector. Commercial-Industrial customers use electricity for space heating and machinery operations.

63

Sales and Customers Data

Number of Minimum Maximum Tax Districts

Residential Electricity Sales (kWh) Number of Residential Electricity Customers Commercial-Industrial Electricity Sales (kWh) Number of CommercialIndustrial Electricity Customers Lighting Electricity Sales (kWh) Number of Lighting Electricity Customers Total Electricity Sales (kWh) Total Number of Electricity Customers

Sum

Mean

Std. Deviation

257

0 1,041,919,003 10,798,844,544 42,018,850 106,304,471

257

0

257

0 2,968,261,212 22,583,320,298 87,872,842 273,398,912

257

0

17,084

202,558

788

1,803

257

0

45,404,964

417,400,526

1,624,127

4,806,710

257 257 257

144,529

1,766,014

6,872

16,669

0 1,433 7,802 30 107 83,415 3,113,865,373 34,669,972,810 134,902,618 357,332,984 3

157,297

1,978,744

7,699

18,480

Table 4.5 Descriptive Statistics for Electricity Customers and Sales

In the natural gas sector, residential customers and sales are larger than commercial-industrial sector customers and sales (Table 4.6). Natural gas is used predominantly for space heating in housing units, and in some cases in cooking.

Sales and Customers Data Residential Natural Gas Sales (mcf) Number of Residential Natural Gas Customers Commercial-Industrial Natural Gas Sales (mcf) Number of CommercialIndustrial Natural Gas Total Natural Gas Sales (mcf) Total Number of Natural Gas Customers

Number of Minimum Maximum Tax Districts

Sum

Mean

Std. Deviation

210

0

7,493,316

80,645,201

384,025

832,184

210

0

71,583

869,042

4,138

8,241

210

0

5,541,700

61,413,043

292,443

646,640

210 210

0 0

5,502 73,324 13,035,016 142,568,540

349 678,898

720 1,432,262

210

1

4,493

8,946

77,085

943,469

Table 4.6 Descriptive Statistics for Natural Gas Customers and Sales 64

Electricity and natural gas sales and numbers of customers are inputs to investment cost models. However, their limited number of available observations is a drawback. In Appendix A, customers and sales data are estimated, when unavailable, using the population and economic census data. Sales data also include information about peak loads, which are the basis for computing peak load factors. For electricity, the peak load factor is defined as the ratio of average hourly sales to maximum hourly sales. For natural gas, it is defined as the ratio of average monthly sales to maximum monthly sales. The peak load factor is an indicator of capacity utilization. The higher the load factor the better the capacity utilization, hence the more efficient the use of resources.

Company Central Hudson Long Island Lighting Company Niagara Mohawk Orange and Rockland

Electricity Load Factor 0.599 0.460 0.626 0.546

Natural Gas Load Factor 0.667 0.549 0.561 0.577

Table 4.7 Company Electricity and Natural Gas Load Factors

4.1.3 Employment Data

The number of employees and wages affect costs. Increasing number of employees and/or wages is likely to increase the labor costs of a firm. Although there are wage variations across different types of employees, the overall average wage per employee can be used to as a proxy for the labor unit cost of a company. 65

Company

Number of Total Wages Wage per Number of Total Wages Wage per Electricity in Electricity Employee Natural Gas in Natural Employee Employees Utility (Electricity) Employees Gas Utility (Gas)

Central Hudson Long Island Lighting Company

1,138

18,915,719

16,622

184

3,041,096

16,528

2,810

49,886,225

17,753

992

20,385,911

20,550

Niagara Mohawk Orange and Rockland

8,081

137,510,452

17,017

1,604

25,479,404

15,885

1,287

20,598,135

16,005

361

5,536,201

15,336

Table 4.8 Company Employees and Average Wages ($)

4.2 Census of Population

Census data are derived from the U.S. Census Bureau’s 1980 Census of Population and Housing, to match the timeline of the company data. The data is available hierarchically at different levels, from the State to the census block levels. Company data, on the other hand, is available at the tax district levels. Tax districts may be cities, villages, and townships. Census data have been processed to be consistent with company data. Census summary level 14, which includes cities, villages, remainders of towns, census designated places (CDP), and towns which do not have village(s) within their boundaries, is detailed enough to derive tax district data. Cities, villages and towns which do not have village(s) within their boundaries are taken as they are, while remainders of towns and CDPs are summed up to obtain town data. Company data is available for tax districts, which are cities, villages and towns. Therefore, the boundaries of tax districts

66

match the boundaries of Census units. The only issue is to create a matching file between tax district codes and Census codes. Census data include detailed information on population and housing characteristics. Demographic information includes total population, number of households, number of houses, median income, median house value, number of rooms, etc. (Table 4.9). Descriptive statistics also provided for the maximum number of observations with sales and customers data, as used in the estimations (Table 4.10). There are also detailed housing characteristics, which will be used in the analysis of electricity and natural gas sales and number of customers, such as the age of the housing stock, the number of stories, the availability of air conditioning, and the types of heating, cooking, and water heating fuels. The census data is available for all the 1619 tax districts that make up the State of New York (Figure 4.7).

67

Number of Minimum Maximum Districts

Census Variable

Sum

Mean

Std. Deviation

Population

1,619

7

2,230,936

17,550,543

10,840

89,307

Households

1,619

3

828,257

6,338,139

3,915

34,935

Housing Units

1,619

3

881,367

6,864,778

4,240

36,964

Aggregate Rooms

1,619

0

3,762,925

33,952,605

20,971

154,235

Median Value of Housing Units

1,619

0

34,934,400 287,167,300 177,373

1,374,966

Median Income of Households

1,619

0

11,718,262 111,006,402

Housing Units with Air Conditioning

1,619

0

427,329

Housing Units Using Utility Gas in House Heating

1,619

0

Housing Units Using Electricity in House Heating

1,619

Housing Units Using Utility Gas in Cooking

68,565

428,632

2,756,094

1,702

17,403

304,134

2,488,526

1,537

12,055

0

31,157

320,798

198

1,091

1,619

0

784,067

4,082,571

2,522

31,938

Housing Units Using Electricity in Cooking

1,619

0

75,745

1,725,174

1,066

3,732

Housing Units Using Utility Gas in Water Heating

1,619

0

332,733

2,717,473

1,678

13,531

Housing Units Using Electricity in Water Heating

1,619

0

28,591

723,877

447

1,092

Low-story houses (1-3)

1,619

0

483,151

4,823,681

2,979

19,036

Mid-story houses (4-6)

1,619

0

331,976

1,084,154

670

12,725

High-story houses (6+)

1,619

0

389,978

795,513

491

10,815

1-5 years

1,619

0

31,768

316,160

195

1,156

Type of Fuel Used in Housing Units

Stories in Housing Units

Age of Housing Units 6-10 years

1,619

0

31,564

478,298

295

1,656

11-20 years

1,619

0

110,199

1,042,533

644

5,009

21-30 years

1,619

0

144,956

1,111,306

686

5,620

30+ years

1,619

0

638,914

3,755,051

2,319

24,587

Table 4.9 Descriptive Statistics for Selected Census Variables

68


Census Variable

Sum

Mean

Std. Deviation

Population

266

16

489,405

5,365,052

21,546

54,239

Households

266

7

150,824

1,791,474

7,195

17,633

Housing Units

266

31

156,470

1,921,471

7,717

18,742

Aggregate Rooms

266

0

2,580,752

30,634,222 123,029

276,487

Median Value of Housing Units

266

28,400

5,402,200

70,240,000 282,088

573,280

Median Income of Households

266

126

994,017

11,134,229

44,716

111,225

Housing Units with Air Conditioning

266

5

111,737

844,054

3,390

9,698

Housing Units Using Utility Gas in House Heating

266

0

128,066

888,392

3,568

10,114

Housing Units Using Electricity in House Heating

266

0

8,314

100,602

404

926

Housing Units Using Utility Gas in Cooking

266

0

114,041

874,069

3,510

10,013

Housing Units Using Electricity in Cooking

266

0

75,745

801,356

3,218

8,120

Housing Units Using Utility Gas in Water Heating

266

0

131,313

958,201

3,848

10,578

Housing Units Using Electricity in Water Heating

266

0

11,756

182,990

735

1,381

Low-story houses (1-3)

266

19

152,600

1,831,052

7,354

18,047

Mid-story houses (4-6)

266

0

3,837

28,758

115

434

High-story houses (6+)

266

0

6,344

26,299

106

581

1-5 years

266

0

2,055

20,002

80

189

6-10 years

266

0

13,747

77,074

310

996

11-20 years

266

0

24,983

163,982

659

1,958

21-30 years

266

0

29,975

335,341

1,347

3,591

30+ years

266

0

134,810

860,065

3,454

10,665

Type of Fuel Used in Housing Units

Stories in Housing Units

Age of Housing Units

Table 4.10 Descriptive Statistics for Selected Census Variables for the Tax Districts with Sales and Customers Data

69

Figure 4.7 Population Distribution in the State of New York

4.3 Economic Censuses

The Economic Censuses include the Census of Retail Trade, the Census of Wholesale Trade, the Census of Manufacturing Industries, and the Census of Service Industries. All data date are for 1982 and include number of establishments, number of employees, and total payroll for each sector. Unfortunately, these data are available only for tax districts with a population over 2,500. There are 280 tax districts with retail and service data, 277 with wholesale data, and 150 with manufacturing data (Table 4.8). Most tax districts have no economic census information due to the population threshold. 70

Figure 4.8 Tax Districts with Economic Censuses

Retail has the largest number of establishments and employees, closely followed by the service sector. The number of employees per establishment is the largest in manufacturing (218). The establishment sizes for the retail, service, and wholesale sectors are very close, and much smaller (Table 4.11 and Table 4.12).

Economic Census Sectors


Sum

Mean

Std. Deviation

Manufacturing Establishments

150

0

66,241 77,151

274

3,945

Manufacturing Employment

150

0 14,438,000 20,526

73

865

Retail Establishments

280

0

12,086 98,742

350

1,290

Retail Employment

280

0

101,126 639,047

2,266

8,356

Service Establishments

280

0

12,002 66,819

237

945

Service Employment

280

0

96,838 510,164

1,809

7,569

Wholesale Establishments

277

0

4,449 22,191

79

349

Wholesale Employment

277

0

42,343 240,229

852

3,652

Table 4.11 Descriptive Statistics for Economic Census Sectors 71

Economic Census Sectors


Sum

Mean

Std. Deviation

Manufacturing Establishments

56

2

1,697

5,540

99

231

Manufacturing Employment

56

5

784

2,825

50

119

2,549

25,142

229

384

23,396 180,339

1,639

3,181

18,675

170

290

20,795 149,756

Retail Establishments

110

4

Retail Employment

110

36

Service Establishments

110

6

Service Employment

110

14

1,361

3,021

Wholesale Establishments

110

1

713

5,692

52

106

Wholesale Employment

110

5

10,011

67,231

611

1,472

1,935

Table 4.12 Descriptive Statistics for Economic Census Sectors for the Tax Districts with Sales and Customers Data

4.4 Geographic Data

Geographic data include information about the natural and built environment. Land use, soil types, slope groups, and street network are components of the geographic data, are all available for the whole State of New York.

4.4.1 Land-Use Data

The source of land-use data is the Geographic Information Retrieval and Analysis System (GIRAS). The U.S. Geological Survey (USGS) provides GIRAS, which contains information on land use and land cover data at the national level. The land-use classification is based on the interpretation of aerial photographies in the 1970s and 1980s. The major categories are: urban or built-up land, agricultural land, rangeland, forest land, water, wetland, barren land, tundra, and perennial snow ice (Figure 4.9). 72

Figure 4.9 Map of Land Use Categories in New York State

73

These categories are further divided into sub-categories. New York State no tundra and perennial snow ice land-uses. The most important land-use group is built-up land, which is expected to affect electricity and natural gas distribution systems the most. Total built-up land covers 3,525 square miles, around 7% of NY State area. Forest land covers 55% of the State (Table 4.13). Total built-up land covers 2,296 square miles for the tax districts with sales and customers data (Table 4.14).


Land Use Type

Sum

Mean

Std. Deviation

Urban and Built-Up Land Residential

1,619

0

100

2,121

1

4

Commercial

1,619

0

16

527

0

1

Industrial

1,619

0

5

104

0

0

Transportation, Communications and Utilities

1,619

0

17

335

0

1

Industrial and Commercial Complexes

1,619

0

3

19

0

0

Mixed Urban

1,619

0

11

135

0

0

Other Urban

1,619

0

8

285

0

1

Agricultural Land

1,619

0

63

15,014

9

11

Rangeland

1,619

0

8

334

0

1

Forest Land

1,619

0

400

27,085

17

30

Water

1,619

0

42

1,364

1

3

Wetland

1,619

0

50

1,090

1

2

Barren Land

1,619

0

7

180

0

0

Total

1,619

48,592

Table 4.13 Descriptive Statistics for Land-use Types (square miles)

74


Land Use Type

Sum

Mean

Std. Deviation

Urban and Built-Up Land Residential

266

0

100

957

3.84

9.31

Commercial

266

0

16

219

0.88

2.03

Industrial

266

0

5

41

0.17

0.52

Transportation, Communications and Utilities

266

0

14

110

0.44

1.24

Industrial and Commercial Complexes

266

0

3

15

0.06

0.31

Mixed Urban

266

0

1

8

0.03

0.12

Other Urban

266

0

5

107

0.43

0.81

Agricultural Land

266

0

50

839

3.37

7.80

Rangeland

266

0

7

20

0.08

0.56

Forest Land

266

0

82

1,196

4.80

11.69

Water

266

0

8

90

0.36

0.96

Wetland

266

0

11

73

0.29

1.23

Barren Land

266

0

7

49

0.20

0.62

Total

266

3,724

Table 4.14 Descriptive Statistics for Land-use Types (square miles) for the Tax Districts with Sales and Customers Data

Built-up land includes 7 sub-groups: Residential, Commercial and Services (commercial areas used for sales of products and services), Industrial (light and heavy manufacturing), Transportation, Communication and Utilities (roads, ports, and processing, treatment, transportation and storage of utilities), Industrial and Commercial Complexes (industrial and commercial land-uses occurring together, such as industrial parks), Mixed Urban or Built-Up Land (individual land-uses that that cannot be separated), and Other Urban or Built-Up Land (urban land-uses such as cemeteries, ski areas, zoos, etc.). 75

The land-use map and the tax district map are overlayed to find out the land-use distribution in each tax district. The percentage of each land use group is calculated by dividing the total area of each land-use by the total area of the tax district. The land use shares must add up to one in each tax district. If district j, and

∑

is the area of land-use i in tax

the correspondence land share, it follows that:

∑

∀

(4.1)

Land use pattern differ across tax districts, depending on the rural or urban character of the area. The following table provides descriptive statistics highlighting these variations.

4.4.2 Slope Data

The topography of the State of New York is derived from the USGS digital elevation model (DEM) files. DEM represents terrain. “This is a raster representation, in which each grid cell records the elevation of the Earth’s surface and reflects a view of terrain as a field of elevation values” (Longley et al., 2005:327). The 7.5-Minute DEM with 30mx30m data spacing has been used. The DEM is converted into raster grid format, where each cell is assigned a slope range. A topographic map of the State is created (Figure 4.10), with 16 slope groups representing slope ranges (Table 4.15).

76

Figure 4.10 Topography Map of New York State

77

Slope Range

Slope Group

Slope Range

Slope Group

Slope Range

Slope Group

0%

1

8%-10%

7

30%-40%

12

0%-0.5%

2

10%-12%

8

40%-50%

13

0.5%-1%

3

12%-15%

9

50%-60%

14

1%-3%

4

15%-20%

10

60%-70%

15

3%-5%

5

20%-30%

11

70%+

16

5%-8%

6

Table 4.15 Slope Groups and Slope Ranges

Raster data is next converted to vector data to enable overlay analysis. Overlaying the slope map and the tax district map shows the slope groups in each geographic unit. The total area of each slope group is divided by the total area of all slope groups in each tax district. The resulting share,

, for land-use i in tax district j, is computed in the

same way as the land-use shares (Equation 4.1). Regrouping of slope groups helps considering good slopes and bad slopes for construction and maintenance of infrastructure investments. Flat sites are easy for transportation and building construction, but they are problematic for drainage. Steep surfaces are hard to work with, due to excavation and transportation facilities, and can be dangerous due to slope instability. It is here necessary to clarify the difference between slope in percent and slope in degrees. Slope in percent is the tangent of the slope in degrees multiplied by 100. The tangent is the ratio of Vertical Distance to Horizontal Distance ( , or the ratio of rise to run. For example, a 14° slope is equal to 25% slope (tan14=0.25). It is 1 unit rise (vertical distance) to 4 units run (horizontal distance) (Figure 4.11). 78

14° slope = 25% slope

Figure 4.11 Relationship between Slope in Degrees and Slope in Percent.

The definition of steepness varies for different building and construction ordinances, but, in general, 0%-5% slopes are considered as flat sites, 5%-10% as low slopes, 10%-20% as moderate slopes, 20%-30% as steep slopes, and above 30% as very steep slopes. Slopes above 30% are not suitable for urban development, and should be conservation land. In fact, extremely steep slopes cover less than one percent of the State of New York, so they are not of concern. Flat sites are the best for infrastructure facilities, but maintenance may be problematic, depending on flood risk and the ease of drainage. Low slopes are considered as most suitable for urban development, and they cover around 23% of the State. Moderate and steep slopes increase construction costs, and these areas cover 17 percent of the total State (Table 4.16).

79

Slopes 0%-5% 0%-0.05% 0.05%-1% 1%-3% 3%-5% 5%-10% 10%-20% 20%-30% 30% and over

Number of Districts 1,619 1,619 1,619 1,619 1,619 1,619 1,619 1,619

Minimum

Maximum

0 0 0 0 0 0 0 0

78 42 111 90 139 93 42 14

Sum 6,166 3,197 12,687 7,517 11,040 6,706 1,176 151

Std. Deviation

Mean 4 2 8 5 7 4 1 0

7 3 10 6 11 9 3 1

Table 4.16 Descriptive Statistics for Areas with Different Slope Types (square miles)

Slopes 0%-5% 0%-0.05% 0.05%-1% 1%-3% 3%-5% 5%-10% 10%-20% 20%-30% 30% and over

Number of Districts

Minimum

266 266 266 266 266 266 266 266

0 0 0 0 0 0 0 0

Maximum 24 42 93 22 31 23 4 1

Sum 261 422 1,124 444 493 229 35 5

Std. Deviation

Mean 1 2 5 2 2 1 0 0

2 4 9 3 4 3 1 0

Table 4.17 Descriptive Statistics for Areas with Different Slope Types (square miles) for the Tax Districts with Sales and Customers Data

4.4.3 Soil Data

The U.S. Department of Agriculture’s State Soil Geographic (STATSGO) database is used for retrieving soil data. STATSGO provides georeferenced digital map that are prepared in 1 by 2 degree topographic quadrangles, and merged to create national coverage at a scale of 1:250,000 (Figure 4.12). 80

Figure 4.12 Soil Types in New York State (there are 169 soil types coded with regard to MUID grids, which are re-grouped according to the workability, corrosivity, average rock depth and water table characteristics, in the analysis)

81

The distribution of soil types in each tax district is calculated with the same method as applied to land use data. A soil map is overlayed with a tax district map to find out the area of each soil type in each tax district. The shares of soil groups are calculated by dividing the total area of each soil type by the sum of all soil groups in a tax district. Let

be the share of soil type i in tax district j. It is computed similarly to land-use

shares (Equation 4.1). However, in contrast to other geographic data (land use, slope and street), soil types are not meaningful without certain characteristics attached. Therefore, soil groups are analyzed according to the soil characteristics that are expected to affect infrastructure investment costs, such as susceptibility to concrete corrosion, susceptibility to steel corrosion, average rock depth, and workability. Soil is composed of disintegrated rock with a combination of organic and inorganic materials in different proportions with regard to different soil types. Gas pipes and underground electricity lines are buried in soil, so soil is expected to have an effect on infrastructure costs. Soil workability and corrosivity characteristics are influential in the investment and maintenance stages. The workability of a soil is defined as a measure of the ease with which a soil is handled and traversed by ordinary construction equipment (Chambers, 1959:14). Soil characteristics, such as structure, consistency, cementation, etc., define the workability conditions of the soil. For instance coarse-grained, soils are considered to be easier to handle in excavation operations than fine-grained soils with high moisture levels. The classification of soil according to particle size, from coarsegrained to fine-grained, are boulders, cobbles, gravel, sand, silt and clay. There is an additional major group named as highly organic soils, such as peat. The same sequence applies to the workability conditions; while the gravelly soils are the more workable and 82

very fine-grained silts and clays are the less workable. Soil types in the data are identified with regard to the American Association of State Highway and Transportation Officials (AASHTO) system. AASHTO classification ranges from A1 to A8. The classes from A1 to A3 are labeled as coarse-grained soils, which can be interpreted as more workable soils. The classes from A4 to A8 are labeled as fine-grained soils, which can be interpreted as less workable soils. AASHTO classes are matched with soil types. Each tax district has a combination of different soil types, and each soil type has a workability value. An average workability is calculated using the percentage of each soil group being attached to a workability value based on AASHTO classes. There are tax districts with a minimum of zero workability, which means that all land is covered by less-workable fine-grained soils. The maximum workability value of 91 implies that none of the tax districts are 100% covered by workable soil (Table 4.18). In the tax districts with sales data, the maximum share of workable land decreases slightly to 90% (Table 4.19).

Soil Factor Workability

Number of Districts 1,619

Minimum 0

Maximum 91

Mean 46

Std. Deviation 16

Mean 55

Std. Deviation 20

Table 4.18 Descriptive Statistics for Soil Workability

Soil Factor Workability

Number of Districts 266

Minimum 0

Maximum 90

Table 4.19 Descriptive Statistics for Soil Workability for the Tax Districts with Sales and Customers Data

83

Corrosivity of a soil is another factor that affects infrastructure systems. Underground corrosion is the deterioration of metals and other materials brought about by the chemical, mechanical, and biological actions of the soil environment. The process of soil corrosion is similar to the galvanic corrosion that happens in galvanic cell (dry cell). In this electrochemical action two dissimilar metals generate galvanic current within an electrolyte. “Most soils contain a mixture of moisture and mineral salts and thereby, satisfy the requirements of a good electrolyte.” (Ibrahim, 1999:10). Underground mains and pipes, which are made from iron, plastic, steel, asbestos cement, reinforced concrete and copper, are subject to corrosion. For instance the copper rod and steel piping immersed in soil creates a galvanic action, while the electrochemical process dissolves metal. Underground corrosion associated with electrical grounding has caused numerous problems in operating electric systems (Zastrow, 1967:237). Palmer (1989), discussing cast iron, states that the corrosivity of soils has increased since the 1960s. Leaks and breakages have been observed due to the weakening effect of corrosion, which has resulted in mechanical failures. In 1965, the first case of stress corrosion cracking (SCC) was reported for an underground gas pipeline. Since then, several occurrences of SCC have been reported in the United States, Australia, Canada, Pakistan, the former Soviet Union, and Iran (Rebak et al, 1996:396). In the STATSGO database, two types of soil corrosion are considered: steel corrosion and concrete corrosion. The former is susceptibility of uncoated steel to corrosion when in contact with the soil, whereas the latter is the susceptibility of concrete to corrosion when in contact with the soil (State Soil Geographic Database, 1995:50). 84

A rating system is available for steel corrosion and concrete corrosion in the STATSGO database. The rating system has the percentages of high, medium, and low corrosion levels for each soil group, which sum up to 100%. In order to create an index for the whole tax district, the high and medium percentages are summed up for each soil type in the tax district. Then, these summary percentages are summed up across all soil types in the tax district, weighted by their areas. The maximum value of 100 indicates that there are tax districts covered 100% with moderate and high corrosive soil, whereas the minimum value of 0 implies existence of districts covered with only low corrosive soil (Table 4.20 and Table 4.21).

Soil Factor Steel Corrosion Concrete Corrosion

Number of Minimum Maximum Districts 1,619 0 100 1,619 0 100

Mean 48 59

Std. Deviation 24 30

Table 4.20 Descriptive Statistics for Steel Corrosion and Concrete Corrosion

Soil Factor Steel Corrosion Concrete Corrosion

Number of Minimum Maximum Districts 266 0 100 266 0 100

Mean 32 76

Std. Deviation 26 29

Table 4.21 Descriptive Statistics for Steel Corrosion and Concrete Corrosion for the Tax Districts with Sales and Customers Data

85

Rock depth and water table are the other soil characteristics that are expected to have effect on investment and maintenance costs. The water table is important for first implementation costs, since a lower water table needs more drainage effort. Rock depth is the depth to bedrock, and it affects excavation costs. The deeper the underground rock level the lesser the excavation costs. Each soil group has a water table level and a rock depth level, and an average is calculated for both water table and rock depth considering the shares of different soil groups in the tax districts. The water table is measured in feet while the rock depth is measured in inches (Table 4.22 and Table 4.23).

Soil Factor Average Rock Depth (inches) Water Table (feet)

Number of Districts Minimum Maximum 1,619 19 60 1,619 2 69

Mean 51 39

Std. Deviation 7 12

Table 4.22 Descriptive Statistics for Average Rock Depth and Water Table

Soil Factor Average Rock Depth (inches) Water Table (feet)

Number of Districts 266 266

Minimum 20 2.3

Maximum Mean 60 52 5.7 3.5

Std. Deviation 6 0.9

Table 4.23 Descriptive Statistics for Average Rock Depth and Water Table for the Tax Districts with Sales and Customers Data

86

4.4.4 Street Data

Street data is derived from Environmental Systems Research Institute (ESRI) street map database. The hierarchical information was obtained from Census Bureau’s TIGER files. Data reflects the information in 1997. Cubukcu (2003) states that the use of 1997 data is reasonable, when the population and growth dynamics of the State of New York is considered over the 1980-1997 period. ‘The growth occurred mostly in New York City through densification, thus with negligible effects on street patterns’ (Cubukcu, 2003:75). New York City is excluded from this study. There are 7 major groups in street data, according to AASHO classification: primary road with limited access or interstate highway; primary road without limited access and state highway; secondary and connecting road, state and county highway; local, neighborhood road, city road and rural road; vehicular trail; ferry crossing; and alley, road for service vehicles (Figure 4.13). Statistics show that most of the roads belong to the fourth group, the local road network, in terms of total length in miles. Descriptive statistics are presented in Table 4.24 and Table 4.25. The total street length in each tax district, the total number of intersections, and the average street segment length (ratio of total street length over total number of intersections) are calculated for each tax district.

87

Figure 4.13 Major Street Types in New York State

88

Road Types Number of Districts Minimum Maximum Primary Roads with Limited Access 1,619 0 78 Primary Roads without Limited Access 1,619 0 18 Secondary and Connecting Roads 1,619 0 55 Local and Neighborhood Roads 1,619 0 2,191 Vehicular Trail 1,619 0 53 Special Roads, Ferry Crossing 1,619 0 49 Other Thoroughfare, Alleys 1,619 0 32 Unknown, Not Classified 1,619 0 40 Total Streets 1,619 2,363 Street Characteristics Number of Intersections 1,619 1 18,569 Street Segment Length 1,619 0 1.2

Sum

Mean Std. Deviation

2,742

2

5

1,769

1

3

11,867

7

8

115,922 1,297

72 1

120 3

917

1

3

154 476 135,143

0 0 84

1 2 131

561,661 476

347 0.3

931 0.2

Table 4.24 Descriptive Statistics for Street Types (miles)

Road Types Number of Districts Minimum Maximum Primary Roads with Limited Access 266 0 39 Primary Roads without Limited Access 266 0 18 Secondary and Connecting Roads 266 0 55 Local and Neighborhood Roads 266 0 2,194 Vehicular Trail 266 0 10 Special Roads, Ferry Crossing 266 0 40 Other Thoroughfare, Alleys 266 0 32 Unknown, Not Classified 266 0 5 Total Streets 266 2,363 Street Characteristics 13 15,114 Number of Intersections 266 0.1 0.5 Street Segment Length 266

Sum

Mean Std. Deviation

668

3

6

355

1

3

1,473

6

9

24,854 72

100 0

214 1

399

2

5

90 11 27,921

0 0 112

2 0 232

185,402

745

1,656

37

0.1

0.1

Table 4.25 Descriptive Statistics for Street Types (miles) for the Tax Districts with Sales and Customers Data 89

4.4.5 Meteorological Data

Meteorological data is derived from the National Oceanic and Atmospheric Administration’s (NOAA) database of Annual Degree Days for 1980. Heating Degree Days (HDD) and Cooling Degree Days (CDD) information are collected for 25 stations in the State of New York (Figure 4.14). Tax districts are assigned to the closest station. The 25 regions are created to assess the variations in meteorological data.

Figure 4.14 Meteorological Stations and Their Catchment Areas

90

The Cortland station has the largest number of tax districts, and the Ogdensburg and Sag Harbor stations cover the smallest numbers within their catchment areas (Table 4.26).

Number of Tax Districts within Minimum Distance to Maximum Distance to the Closest Tax District the Closest Tax District Station Name Station Catchment Area Little Valley 89 3.8 65.52 Cortland 177 2.78 72.49 Liberty 35 8.42 45.52 Port Jervis 16 5.79 17.39 Sag Harbor 12 3.78 24.9 Mineola 112 1.33 28.62 Patchogue 19 7.83 19.73 Albany 99 2.96 38.26 Glens Falls 73 3.4 78.13 Stockbridge 73 3.9 44.83 Poughkeepsie 36 4.26 36.46 Highland Falls 79 2.76 23.96 Gloversville 69 2.94 42.46 Utica 86 3.42 50.62 Canton 33 8.28 54.82 Massena 58 7.03 80.81 Ogdensburg 12 5.74 26.04 Batavia 160 4.41 67.65 Buffalo 45 5.02 32.27 Fredonia 46 2.35 33.45 Lockport 25 5.35 19.8 Oswego 71 3.15 48.47 Rochester 56 5.85 49.08 Watertown 73 3.61 48.08 Syracuse 65 3.65 44.06

Table 4.26 Stations and Statistics on Tax Districts within their Catchment Areas1

1

The closest tax district excludes the district where the station is located.

91

The annual HDD and CDD are directly related to the amounts of energy used for heating and cooling, respectively, and are calculated by using the difference between a base temperature and the actual temperature on a given day. According to the National Climatic Data Center (NCDC) of the National Oceanic and Atmospheric Administration (NOAA), the base temperature is 65 Degrees Fahrenheit in the US (NOAA, Monthly Station Climate Summaries). The annual HDD and CDD values are the sums of the daily HDD and CDD values. The average monthly HDD and CDD are obtained by dividing the annual totals by 12. Descriptive statistics on HDD and CDD are provided in Table 4.27.

Meteorological Factors Heating Degree Days Annual Sum Heating Degree Days Peak Load Heating Degree Days Monthly Average Cooling Degree Days Annual Sum Cooling Degree Days Peak Load Cooling Degree Days Monthly Average

Number of Districts

Minimum

Maximum

Mean

Std. Deviation

1,619

5,736

8,774

7,227

737

1,619

2.0

2.4

2.2

0.1

1,619

478

731

602

61

1,619

324

1,158

603

182

1,619

3.9

5.2

4.6

0.3

1,619

27

97

50

15

Table 4.27 Descriptive Statistics for Meteorological Data

92

Meteorolical Factors Heating Degree Days Annual Sum Heating Degree Days Peak Load Heating Degree Days Annual Average Cooling Degree Days Annual Sum Cooling Degree Days Peak Load Cooling Degree Days Annual Average

Number of Districts

Minimum

Maximum

Mean

Std. Deviation

266

5,736

8,774

6,326

685

266

2.0

2.4

2.3

0.1

266

453

731

526

58

266

324

1,158

814

190

266

3.9

5.2

4.5

0.3

266

27

97

68

16

Table 4.28 Descriptive Statistics for Meteorological Data for the Tax Districts with Sales and Customers Data

The geographic variations in total annual HDD values (Figure 4.15) indicate that the northern part of the State is the coolest area, whereas the warmest areas tend to concentrate in more populated and dense urban areas, such as Syracuse and New York City. The CDD annual totals (Figure 4.16) make up a map that is more or less the reverse of the HDD map. The geographic variations in the HDD and CDD annual sums can be linked to urban heat effect, since higher densities result in less heating and more cooling needs.

93

Figure 4.15 Variations of Annual Heating Degree Days across the State of New York

Figure 4.16 Variations of Annual Cooling Degree Days across the State of New York 94

CHAPTER 5 MODELS OF AGGREGATE CAPITAL INVESTMENTS IN ELECTRICITY AND NATURAL GAS DISTRIBUTION

This chapter presents capital cost models and economies of scope analyses for aggregate electricity and natural gas distribution plants. The capital cost models consider the multi-product character of electricity and natural gas distribution utilities. In addition to cost estimation, elasticity and economies of scale analyses are also presented. The analysis of economies of scope assesses whether combined service provision is cost advantageous. All models include various locale-specific variables showing the impacts of site specific and geographic variables.

5.1 Aggregate Capital Investment Cost Functions

5.1.1 Overview

A market analysis was carried out to expand the sample of market communities to potentially cover the whole state of New York in terms of numbers of customers and sales in each sector (Appendix A). However, this larger dataset failed to yield reasonable estimates for the aggregate capital investment costs. Hence, all forthcoming analyses are 95

restricted to the communities (tax districts) where actual electricity and natural gas sales and numbers of customers are available. The estimated cost functions are therefore considered as valid for the size ranges of these communities, but may not apply to smaller communities as well as to much larger ones (e.g. New York City). Costs are considered functions of the numbers of customers and sales in different sectors (residential, commercial-industrial, and lighting), urban site-specific variables such as density, environmental factors such as soil type and topography, and company specific variables such as load factor and wage per employee. As discussed in Chapter 3, the price of capital is not considered, because the necessity data is not available. Generally:

Ci = F (Qij, SITE, G, COMPi)

(5.1)

where Ci = Aggregate capital investment costs in sector i (gas or electricity) Qij = Quantity of output j (e.g. residential sales) in sector i SITE = Site-specific variables G = Geographic variables COMPi = Company-specific variable in sector i.

Guldmann (1989) suggests that there is no agreed-upon theory regarding the functional form of the capital cost function, and uses linear, log-log and Box-Cox regressions. Linear, log-log and Box-Cox estimated functions are compared, and the BoxCox form appears to be superior in all estimations, using the log-likelihood ratio. 96

The log-log function is defined as

(5.2)

The Box-Cox function is defined as

where

), or:

(5.3)

(5.4)

The elasticity of cost (

) in the log-log form is the coefficient (

of the

corresponding independent variable, and is constant. The cost elasticity in the case of the Box-Cox equation, on the other hand, is a function which varies with different input/output values:

(5.5)

Both electricity and natural gas utilities are multi-product firms, since they provide different outputs to different customer groups: residential, commercial-industrial, and lighting (electricity only). Economies of scale in multi-product firms are measured by 97

ray economies of scale (εR), which are here computed as the sum of the elasticity values for the different outputs:

∑

(5.6)

Elasticity is a measure (%) of the change in the dependent variable resulting from a 1% change in an independent variable. Economies of scale take place when diseconomies when

, and constant returns to scale when

,

Note that

Baumol at al. (1982) define ray economies of scale as the inverse of the sum of cost elasticites (

∑

. Generally,

economies of scale consider output expansion while

holding density constant, that is, while expanding the area of service at the same rate as the output, whereas economies of density, εD, consider output expansion within a fixed area, hence densification. Economies are achieved through densification if

:

εD = εR + εDENS

(5.7)

Population density is a proxy for network size per customer (e.g., miles of lines per customer), because data on mileages of lines (electricity or gas) are not available. An increasing population density is therefore taken as equivalent to adding customers (and sales) to a fixed-length network. While the firm has no direct control over population density, it can certainly influence it with policies of network expansion, particularly the differentiated pricing of this expansion. 98

5.1.2 Cost Model for Aggregate Electricity Distribution Investment

5.1.2.1 Model Estimates

The aggregate electricity distribution plant comprises overhead and underground lines, conduits, services, transformers, poles, and street lighting equipment. Electricity distribution investment costs (CE) are considered as a function of outputs, input prices, and site specific characteristics. Industry outputs are the numbers of customers and sales in three sectors: residential, commercial-industrial, and lighting. However, lighting customers and sales, and commercial-industrial customers turned out to be insignificant. The site variables that turned out to be significant are: population density, soil corrosivity, and the number of street/road intersections. Density is the site-specific variable that has been widely used in the literature, and area density is used in this model. Increasing soil corrosivity is expected to increase capital costs. The number of intersections is related to urban form: more intersections lead to higher costs due to the need for special crossing measures and maintenance costs. The load factor, which is an indicator of capacity use, is another variable that turned out to be significant. The distribution capacity system must be designed to accommodate peak flows. A higher load factors implies a more efficient utilization of capacity, hence lower costs based on a given annual load. The selected model is:

CE = F(NRE, SRE, SCIE, DENSA, SOILCORR, INTR, LFE) where 99

(5.8)

CE = Electricity distribution investment costs ($) NRE = Number of residential electricity customers (#) SRE = Residential electricity sales (kWh) SCIE = Commercial-Industrial electricity sales (kWh) DENSA = Area Density (Population/Total square miles of tax district) SOILCORR = Soil Corrosivity (%) INT = Number of street intersections (#) LFE = Electricity load factor (Average hourly kWh sales/Hourly kWh peak load)

Descriptive statistics for the variables in Eq. (5.8) are presented in Table 5.1.

Variable CE ($) NRE (#) SRE (kWh) SCIE (kWh) DENSA (pop./sq.m.) SOILCORR (%) INT (#) LFE

Minimum 13,968 20 167,697 1,760 14 2.78 13 0.46

Maximum 319,335,460 144,529 1,041,919,003 2,968,261,212 103,058 99.9 15,114 0.63

Mean 11,246,511 6,821 43,503,227 90,696,852 3,517 78.14 751 0.53

Std. Deviation 29,383,157 16,662 107,788,769 277,374,196 7,202 29 1,659 0.07

Table 5.1 Descriptive Statistics for the Aggregate Electricity Distribution Investment Cost Model (n=241)

The regression results are presented in Table 5.2. The Box-Cox form is superior to the log-log form, in terms of log-likelihood values. All variables are statistically significant, with the expected signs. 100

Coefficient Constant NRE SRE SCIE DENSA SOILCORR INTR LFE

Models Log-log 2.556 (3.13)b 0.339 (3.81) 0.461 (5.74) 0.126 (4.87) -0.176 (-6.37) 0.136 (2.17) 0.090 (1.87) -0.736 (-2.24)

Box-Cox(λ, θ)a 3.904 (2.52) 0.853 (4.34) 0.331 (4.24) 0.130 (4.75) -0.351 (-5.70) 0.407 (2.19) 0.254 (2.00) -3.400 (-2.37)

λ

0.090 (0.003)c 0.093 (0.010)

θ R2

0.927

0.933

-3741.02

-3736.49

H0: θ=λ=0, Chi-sq=9.07d, H0: θ=λ=1, Chi-sq=740.28, a Selected model b t-statistics in parentheses c p-value in parentheses d Log-likelihood test results

p>Chi-sq = 0.003 p>Chi-sq = 0.000

Log-likelihood

Table 5.2 Aggregate Electricity Distribution Cost Function Estimates (n=241)

Cost elasticities at the sample mean are presented in Table 5.3. The number of residential customers has a strong impact on total costs, and residential sales increases the costs almost three times more than commercial-industrial sales. Investment costs increase with soil corrosion and the number of intersections, but decrease with area density and load factor. The negative coefficient of the density variable can be explained by savings 101

from capital and labor input in dense areas when compared to a dispersed pattern. The load factor, εLFE, has the highest absolute elasticity among all variables. A 1% increase in load factor would decrease costs by 0.7% at the sample mean. A 1% increase in commercial industrial sales (εSCIE), area density (εSCIE), soil corrosivity (εSOILCORR), and the number of intersections (εINTR) would increase the costs by less than 0.2%. The effects of residential customers and sales are relatively close to each other, with a 1% increase in any of these variables resulting in almost a 0.4% increase in costs.

Elasticity

εNRE εSRE εSCIE εDENSA εSOILCORR εINTR εLFE

Sample Mean 0.415 0.352 0.147 -0.161 0.133 0.101 -0.709

Table 5.3 Aggregate Electric Distribution Cost Elasticities at the Sample Mean (n=241)

5.1.2.2 Economies of Scale

An electricity distribution utility is considered as a multi-product firm, providing service to various customers: residential, commercial-industrial, and lighting. Ray economies of scale in the case of the model in Table 5.2 are measured by.

εCE = εNRE + εSRE + εSCIE

(5.9) 102

Using the Box-Cox function in Table 5.2, Equation (5.9) becomes:

[

]

(5.10)

Ray economies of scale are calculated at the sample mean, as well as for each individual observation (tax districts). The value of 0.914 for εCE at the sample mean represents slight economies of scale achieved through system expansion at constant density. The lowest, highest, and average values of economies of scale for individual observations are 0.837, 0.996, and 0.907 respectively. The variations of

across the

241 observations are presented in Figure 5.1. All points are below but fairly close to 1, which points to slight economies of scale in most districts.

1.2 1

εCE

0.8 0.6 0.4 0.2 1 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171 181 191 201 211 221 231 241

0

Observations

Figure 5.1 Point Elasticities of Scale across 241 Observations (tax districts).

103

Varying output levels are expected to affect economies of scale. The outputs are the numbers of residential customers, residential sales, and commercial-industrial sales. All of them are measured in different units: actual numbers, kilowatt hours, and thousand cubic feet. A parameter, k, is used to represent residential and commercial-industrial sales, which are assumed to expand at the same rate, while keeping the residential customer size ZRE (ratio of residential sales to number of customers) constant. Therefore, the number of residential customers also expands at the same rate. For residential sales, commercial-industrial sales and the number of residential customers, we have: (1) SRE=k*SREmean=k*43,503,227, (2) SCIE=k*SCIEmean =k*90,696,852, and (3) NRE=SRE/ZRE=k*43,503,227/ZR, respectively. Different output levels are considered for the minimum, mean, and maximum values of ZRE. Figure 5.2 shows the variations of economies of scale versus k, for ZREmin (3351), ZREmean (7572), and ZREmax (25689) values. Whatever ZRE, economies of scale decline (εCE increases) with an increasing output, suggesting that larger markets (i.e. larger cities) provide less opportunities for economies of scale. For a given market size (k fixed), the larger ZRE the larger the economies of scale. When ZRE increases in a given market, the number of residential customers decreases. It is expensive to serve areas where the number of customers is at minimum, therefore there are more opportunities to decrease costs by hooking up new customers and expanding sales. When the outputs increase by 1000 percent, εCE becomes very close to 0.924 for ZREmin. The elasticity curve of ZREmin has a horizontal asymptote, at εCE=0.961: no matter how high the increase in outputs, economies of scale do not exceed 0.961 and never turn into diseconomies of scale. For a more detailed mathematical analysis of the εCE function, see Appendix B. 104

Figure 5.2 Economies of Scale, εCE, versus Output Parameter, k.

Site-specific, geographic, and company-specific variables affect costs, and therefore are expected to affect economies of scale. Changes in the economies of scale curves are assessed for the minimum, mean, and maximum levels of the following variables: area density, soil corrosivity, number of intersections, and electricity load factor. The mean customer size, ZRE= 7572, is used in all cases. The value of εCE never exceeds 1. The curve of εCE versus k is almost horizontal for the maximum area density, which suggests that, at high densities, the cost elasticity is almost independent of the level of output, with a constant value of approximately 0.97. The minimum density yields much higher economies of scale, with εCE values ranging between 0.80 and 0.88. Soil corrosivity has an outlier minimum value of 2.78 percent, which leads to an almost horizontal curve at εCE=0.95. At the mean and maximum corrosivity levels, the elasticity curves are very close to each other, between 0.90 and 0.92. Economies of scale vary from 105

0.89 to 0.95 when considering the minimum and maximum numbers of street intersections. A higher load factor leads to lower economies of scale, around εCE=0.94. Higher load factors provide fewer opportunities to increase capacity utilization. When the capacity is better utilized, there is less room left for cost reductions, while adjustments in the existing system become more expensive.

(a) Area Density (DENSA) Figure 5.3 Economies of Scale, εCE, versus Output Parameter, k, for Different Levels of the Site-specific, Geographic, and Company-specific Variables: (a) Area density, (b) Soil corrosivity, (c) Number of street intersections, (d) Electricity load factor

106

Figure 5.3 continued

(b) Soil Corrosivity (SOILCORR)

(c) Number of street intersections (INT)

107


(d) Electricity load factor (LFE)

5.1.2.3 Economies of Density

The economies of density, εDE, represent the effects of both service densification and expansion, with:

εDE = εCE + εDENSA

(5.11)


[

]

108

(5.12)

The value of εDE at the sample mean is 0.754, pointing to higher economies resulting from densification. Economies of density are also calculated for individual observations, with minimum, maximum and average values at 0.664, 0.794, and 0.737, respectively. The variations of economies of density for the 241 observations are presented in Figure 5.4. Densification leads to higher economies, as compared to solely service expansion.

0.9 0.8 0.7

εDE

0.6 0.5 0.4 0.3 0.2 0.1 1 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171 181 191 201 211 221 231 241

0

Observations

Figure 5.4 Point Elasticities of Density across 241 Observations (tax districts).

The variations in economies of density εDE with the output parameter, k, are analyzed for the minimum, mean, and maximum values of customer size, ZRE (Figure 5.5). For any value of ZRE, economies of density decline (εDE increases) with an increasing output. For a given market size (k fixed), the larger ZRE the larger the 109

economies of density. Such areas, with a smaller numbers of customers, probably provide more opportunities for savings by hooking up new infill customers. The elasticity curve of ZREmin has a horizontal asymptote at εDE=0.859, which can be compared to the limit value of 0.961 for εCE.

Figure 5.5 Economies of Density, εDE, versus Output Parameter, k.

Low-density areas have the highest economies of density (Figure 5.6), because these areas are very expensive to serve, and therefore increasing their densities by adding customers through “infill” is likely to significantly reduce unit costs, hence high levels of εDE. Higher densities decrease costs, but provide less opportunities for larger economies of scale. Areas with more street intersections also benefit from densification in a significant way: a larger number of street intersections may provide more flexibility in 110

extending the network to connect new customers. A lower soil corrosivity and a higher electricity load factor, in contrast, decrease the levels of εDE. The maximum soil corrosivity leads to the largest economies of density: expansion of lines through densification require shorter lines, less exposed to corrosion, hence the higher economies. A higher load factor is an indicator of higher levels of capacity utilization, which may limit the potential gains through expansion with densification.

(a) Area density (DENSA) Figure 5.6 Economies of Density, εDE, versus Output Parameter, k, for Different Levels of the Site-specific, Geographic, and Company-specific Variables: (a) Area density, (b) Soil corrosivity, (c) Number of street intersections, (d) Electricity load factor

111


(b) Soil corrosivity (SOILCORR)


112


(d) Electricity load factor (LFE)

5.1.3 Cost Model for Aggregate Natural Gas Distribution Investment

5.1.3.1 Model Estimation

The aggregate natural gas distribution plant comprises mains, services, measurement and regulation stations, and structure and improvements. Natural distribution investment costs (CG) are considered function of outputs, input prices, and site specific characteristics, as in the case of electricity distribution. Industry outputs are the numbers of customers and sales in two sectors: residential and commercial-industrial. The number of commercial-industrial customers however, turned out to be insignificant. The built-up area, which is a proxy for urban development, turned out to be significant. 113

Another significant urban-related variable is the number of intersections, expected to increase costs in a way similar to electricity distribution. Soil workability is the geographic variable that has a significant effect on costs. Natural gas investments are largely underground, and therefore an increase in the percentage of workable soil is expected to decrease costs. The labor-intensive character of underground work makes the wage per employee another significant variable affecting the total cost. The selected model is:

CG = F(NRG, SRG, SCIG, ABLTP, SOILWORK, INTR, WPEG) where CG = Natural gas distribution investment costs ($) NRG = Number of residential gas customers (#) SRG = Residential gas sales (mcf) SCIG = Commercial-Industrial gas sales (mcf) ABLTP = Total built-up area (square miles) SOILWORK = Workable soil (%) INTR = Number of street intersections (#) WPEG = Average wage per employee in natural gas sector ($)

Descriptive statistics for the above variables are presented in Table 5.4.

114

(5.13)

Variable CG ($) NRG (#) SRG (kWh) SCIG (mcf) ABLTP (sq miles) SOILWORK (%) INTR (#) WPEG ($)

Minimum 34,874 4 8 4 0.04 1.56 13 15,336

Maximum 108,274,446 71,583 7,493,316 5,541,700 134 86 15,114 20,550

Mean 7,656,032 4,471 411,433 311,081 6.51 58 851 18,143

Std. Deviation 14,548,161 8,515 859,346 667,499 14.45 18 1,821 2,388

Table 5.4 Descriptive Statistics for the Total Natural Gas Investment Cost Model (n=190)

The regression results are presented in Table 5.5. The Box-Cox form is superior to the log-log form, using the log-likelihood ratio-test. Urban related variables – built-up area and the number of intersections – have positive effects on costs, as expected. The larger the built-up area, the larger the investment costs. Intersections make physical work more complex and necessitate more capital input and labor work. Workable soil, on the other hand, makes workmanship easier, and therefore has a negative effect on costs. An increase in wage per customer increase total costs directly. Cost elasticities at the sample mean are presented in Table 5.6. The average wage gas employee (εWPEG) has the highest elasticity among all the variables. When the average wage increases by 1%, the cost increases by 1.26% at the sample mean. Residential sales (εSRG) and soil workability (εSOILWORK) have the second and third highest elasticity levels, but with opposite signs. The elasticities of the number of residential customers (εNRG), commercial-industrial sales(εSCIG), built-up area (εABLTP), and the number of intersections (εINTR) are close to each other: a 1% increase in any of these variables results in less than a 0.2% increase in costs. 115

Coefficient Constant NRG SRG SCIG ABLTP SOILWORK INTR WPEG

Models Log-log -6.627 (-2.09)b 0.148 (2.98) 0.422 (9.98) 0.083 (3.14) 0.140 (2.1) -0.297 (-3.28) 0.205 (2.54) 1.466 (4.41)

Box-Cox(λ, θ)a -123.137 ( -4.49) 0.672 ( 2.72) 1.437 (13.23) 0.172 (2.07) 2.183 (2.09) -4.047 (-4.16) 1.354 (2.61) 5.920 (5.27)

λ

0.177 (0.000)c 0.206 (0.000)

θ R2

0.914

0.941

Log-likelihood -2911.77 H0: θ=λ=0d, Chi-sq=43.16, H0: θ=λ=1, Chi-sq=436.22, a Selected model b t-statistics in parentheses c p-value in parentheses d Log-likelihood test results

-2890.2 p>Chi-sq = 0.000 p>Chi-sq = 0.000

Table 5.5 Total Natural Gas Distribution Cost Function Estimates (n=190)

Elasticity

εNRG εSRG εSCIG εABLTP εSOILWORK εINTR εWPEG

Sample Mean 0.112 0.531 0.060 0.114 -0.311 0.167 1.259

Table 5.6 Aggregate Natural Gas Distribution Cost Elasticities (n=190) 116

5.1.3.2 Economies of Scale

Ray economies of scale are measured by the effect of increases in sales and numbers of customers at constant density. The built-up area variable capture area expansion, and increasing it keeps density constant. It follows that:

εCG = εNRG + εSRG + εSCIG + εABLTP

(5.14)


[

]

(5.15)

εCG has a value of 0.817 at the sample mean, pointing to economies of scale. The lowest, highest, and average values of εCG computed for all the individual observations (190) are 0.689, 0.841, and 1.02, respectively. The variations of εCG are presented on Figure 5.7. There are three observations (tax districts) for which εCG>1: 1.02, and twice 1.01. The common characteristics of these three tax districts are very small built-up areas and very small numbers of street intersections, as compared to the sample mean values (Table 5.7).

117

1.2 1

εCG

0.8 0.6 0.4 0.2

1 9 17 25 33 41 49 57 65 73 81 89 97 105 113 121 129 137 145 153 161 169 177 185

0

Observations

Figure 5.7 Point Elasticities of Scale across 190 Observations (tax districts).

Name

NRG

SRG

SCIG

ABLTP

SOILWORK

INT

WPEG

Haverstraw Village Wappingers Falls Village Grand View on Hudson Village

2,192

301,396

174,993

0.04

69.9

13

15,336

1.02

2,533

8,242

6,163

0.11

77.55

37

16,528

1.01

156

30,797

983

0.18

48.24

11

15,336

1.01

Table 5.7 Statistics for the Tax Districts with Diseconomies of Scale: εCG>1

The variations in economies of scale with output are analyzed in a way similar to the analysis of electricity economies of scale, for the minimum, ZRGmin, the mean, ZRGmean, and the maximum, ZRGmax, values of the average residential customer size. The parameter k is derived for residential sales, commercial-industrial sales and the number of residential customers as follows: (1) SRG=k*SRGmean =k*411,433, (2) SCIG=k*SCIGmean 118

=k*311,081, and (3) NRG=SRG/ZRG=k*411,433/ZRG. The εCG curves are presented on Figure 5.8. Whatever ZRG, economies of scale decline (εCG increases) with an increasing output, in a way similar to the economies of scale in electricity. Larger markets provide less opportunities for economies of scale. If the market size is fixed (k fixed), the smaller ZRG the smaller the economies of scale. Increasing values of ZRG results in decreases in the number of residential customers. Areas with less customers are more expensive to serve, hence provide more opportunities for cost reductions with service expansion. The limit value for the elasticity with ZRGmin=3 is 0.859, which indicates that the gas distribution system does not experience diseconomies for any customer size value when all the other variables are at their sample mean values, However, the results in Table 5.7 show that there may be outlier local conditions when diseconomies of scale actually take place. For a more detailed mathematical analysis of the εCG function, see Appendix B.

Figure 5.8 Economies of Scale, εCG, versus Output Parameter, k. 119

Changes in economies of scale are analyzed for the minimum, mean, and maximum levels of built-up area, soil workability, number of intersections, and wage per employee, for the mean residential customer size level ZRG= 106.2 (Figure 5.9). In all cases, the value of εCG does not exceed 1. The curves for built-up area and number of intersections are similar, but with different ranges. Varying the built-up area yields εCG values ranging between 0.70 and 0.90, while varying the number of street intersections yield εCG values ranging between 0.68 and 0.94. The minimum built-up area and the minimum number of street intersections yield the lowest economies of scales (highest curves). Expanding output while simultaneously expanding territory under such conditions may be expensive and provides little room for economies of scale. Soil workability has an outlier minimum value of 1.56. At the mean and maximum soil workability levels, the elasticity curves are almost horizontal and very close to each other between 0.80 and 0.86. More workable soil leads to lower costs, and it becomes hard to decrease costs further, whereas in areas with less workable soil the costs are high, thus cost reductions are more achievable. The minimum wage level yields a lower level of εCG, around 0.86. Lower wages may imply more labor substituted for capital, hence lesser opportunities for capital economies of scale.

120

(a) Built-up area (ABLTP)

(b) Soil workability (SOILWORK) Figure 5.9 Economies of Scale, εCG, versus Output Parameter, k, at Different Levels of the Site-specific, Geographic, and Company-specific Variables: (a) Built-up area, (b) Soil workability, (c) Number of street intersections, (d) Wage per employee.

121



(d) Wage per employee (WPEG)

122

5.1.3.3 Economies of Density

The economies of density, εDG, represent the effects of both service densification and expansion. The elasticity of built-up area is removed from Equation (5.16) to capture the effect of densification, with:

εDG = εCG - εABLTP = εNRG + εSRG + εSCIG

(5.16)


[

]

(5.17)

The value of εDG at the sample mean is 0.703. Economies of density are also calculated for individual observations (tax districts), with minimum, maximum and average values at 0.512, 0.953, and 0.717, respectively. The variations of εDG for the 190 observations are illustrated on Figure 5.10. Economies of density never exceed 1.

123

1.2 1

εD

0.8 0.6 0.4 0.2

1 9 17 25 33 41 49 57 65 73 81 89 97 105 113 121 129 137 145 153 161 169 177 185

0

Observations

Figure 5.10 Point Elasticities of Density across 190 Observations (tax districts).

The variations in economies of density, εDG, with the output parameter, k, are analyzed for the minimum, mean and maximum values of residential customer size, ZRG (Figure 5.11). The maximum and mean average customer sizes yield economies of density levels very close to each other. Economies of density are greater for higher customer sizes, because larger markets (i.e., larger cities) provide more opportunity for economies of density. In a given market (k fixed), increasing ZRG means decreasing the number of customers, which probably provides savings through hook-up of new infill customers.

124

Figure 5.11 Economies of Density, εDG, versus Output Parameter, k.

The economies of density for different levels of site-specific, geographic, and company-specific variables are presented in Figure 5.12. In all cases, the mean customer size, ZRGmean, is used. Larger built-up areas benefit more from densification than smaller built-up areas, possibly because of better hook-up opportunities for infill customers. The rankings of the elasticity curves for soil workability, number of street intersections, and wage rate are similar to those observed in the case of expansion without densification (Figure 5.9), and the same interpretations of the results apply here.

125

(a) Built-up area (ABLTP)

(b) Soil workability (SOILWORK) Figure 5.12 Economies of Density, εD, versus Output Parameter, k, at Different Levels of the Site-specific, Geographic, and Company-specific Variables: (a) Built-up area, (b) Soil workability, (c) Number of street intersections, (d) Wage per employee.

126



(d) Wage per employee (WPEG)

127

5.2 Multi-Utility Cost Analysis

The presence of economies of scope in dual electricity and natural gas distribution firms is examined in this section. Two hundred and forty six (246) tax districts are used in this analysis: 60 districts where a company serves only electricity, and 186 where a company serves both electricity and gas. There are 20 districts for which only gas sales are available. However, there is an electricity distribution plant in these districts, but electricity sales were not provided in the utility (Central Hudson) annual report. Hence, these 20 districts cannot be considered in this analysis. Multi-utility investment cost is modeled as a function of total electricity and natural gas sales. Following Mayo (1984), a dummy variable was added to distinguish electricity only (DE) firm, but DE turned out to be insignificant and was dropped from the final model. The number of intersections reflect the effect of urban form on costs. The share of old housing stock (houses more than 40 years old) is significant and can be considered as a proxy for concentration of neighborhoods close to city center. Old housing stock is expected to increase costs due to their central location, which may call for special pavement in worn and narrow streets, and more maintenance. The selected model is:

C = F(STE, STG, INTR, AGE_P40)

(5.18)

where STE = Total electricity sales (kWh) STG = Total natural gas sales (mcf) INTR = Number of street intersections (#) 128

AGE_P40 = Share of 40+ year old houses (%)

Descriptive statistics for above variables are presented in Table 5.8.

Variable C ($) STE (kWh) STG (mcf) INTR (#) AGE_P40 (%)

Minimum 3,236 83,415 142 13 0.003

Maximum 319,335,460 3,113,865,373 13,035,016 15,114 0.90

Mean 16,713,214 135,749,892 719,058 742 0.37

Std. Deviation 37,500,000 364,172,187 1,497,039 1,665 0.21

Table 5.8 Descriptive Statistics for the Multi-utility Investment Cost Model (n=246)

The regression results show that the Box-Cox form is superior to the log-log one, based on the log-likelihood test results (Table 5.9). All the variables are significant, with the expected signs.

129

Coefficient Constant STE STG INTR AGEP40

Models Log-log 2.690 (5.31)b 0.643 (15.74) 0.028 (7.59) 0.298 (5.71) 0.193 (3.47)

Box-Cox(λ, θ)a 17.179 (5.28) 0.390 (14.35) 0.389 (10.75) 4.095 (10.09) 8.249 (4.16)

0.849

0.211 (0.000)c 0.225 (0.000) 0.919

λ θ R2

Log-likelihood -4014.092 H0: θ=λ=0d, Chi-sq=118.84, H0: θ=λ=1, Chi-sq=885.00, a Selected model b t-statistics in parentheses c p-value in parentheses d Log-likelihood test results

-3954.67 p>Chi-sq = 0.000 p>Chi-sq = 0.000

Table 5.9 Multi-utility Cost Function (n=246)

Economies of scope exist when the total cost of provision of the two commodities separately is higher than the cost of the combined service, with.

(5.19)

In order to measure the extent of economies of scope, the scope score

S=

is calculated:

(5.20) 130

Positive values of S indicate the presence of economies of scope, whereas negative ones indicate diseconomies. Values of S are calculated for different combinations of electricity and natural gas outputs, STE and STG, while using the mean values of all the other variables (Table 5.10). The range between the minimum and the maximum values of STE and STG, [83 – 3,113,865] and [0.14 – 13,035], respectively, are divided into 10 equal intervals. S values are measured for each output combination, and appears to vary between 0.234 and 0.771, providing evidence for economies of scope. Economies of scope decrease with increasing electricity and natural gas outputs (Table 5.10 and Figure 5.13). The highest level of economies of scope (0.771) is obtained when both sectors’ outputs are minimal (83 Mwh and 0.14 mmcf), whereas the lowest level of economies of scope (0.234) is obtained when both sectors’ outputs are maximal (3,113,865 Mwh and 13,035 mmcf). The higher economies of scope in smaller markets may be related to easier coordination of investments in both distribution systems.

ELECTRICITY (1000 kwh)

NATURAL GAS (1000 mcf)

83 346,061 692,039 1,038,018 1,383,996 1,729,974 2,075,952 2,421,930 2,767,908 3,113,865 0.14 0.771

0.382

0.349

0.330

0.317

0.307

0.299

0.292

0.286

0.281

1,448 0.579

0.328

0.303

0.289

0.279

0.271

0.265

0.260

0.255

0.251

2,896 0.554

0.320

0.296

0.283

0.273

0.265

0.259

0.254

0.250

0.246

4,345 0.539

0.315

0.292

0.279

0.269

0.262

0.256

0.251

0.247

0.243

5,793 0.527

0.311

0.289

0.276

0.266

0.259

0.253

0.249

0.244

0.241

7,241 0.519

0.308

0.286

0.273

0.264

0.257

0.251

0.247

0.243

0.239

8,690 0.511

0.305

0.284

0.271

0.262

0.255

0.250

0.245

0.241

0.237

10,138 0.505

0.303

0.282

0.269

0.260

0.254

0.248

0.244

0.240

0.236

11,586 0.500

0.301

0.280

0.268

0.259

0.252

0.247

0.242

0.238

0.235

13,035 0.495

0.299

0.279

0.266

0.258

0.251

0.246

0.241

0.237

0.234

Table 5.10 Scope Scores for Different Levels of Electricity and Natural Gas Outputs

131

Figure 5.13 Scope Scores for Different Levels of Electricity and Natural Gas Outputs.

Values of

are calculated for the same output levels, as in Table 5.10, while

using the minimum and maximum values of site-specific variables: number of street intersections and the share of houses 40+ years old.

scores for 4 combinations of

outputs and at the minimum/maximum values of the site-specific variables are presented in Table 5.11. The complete sets of score values are presented in Appendix C, and graphical representations of the scope surfaces are presented on Figure 5.14.

Site-specific variables INTR AGEP40

Output levels (electricity ; natural gas) (83 ; 0.14)

(83 ; 13,035)

(3,113,865 ; 0.14)

(3,113,865 ; 13,035)

min

0.471

0.206

0.094

0.075

max

0.868

0.656

0.433

0.373

min

0.709

0.414

0.221

0.181

max

0.786

0.516

0.299

0.249

Table 5.11 Scope Values for Different Levels of Site-specific Variables and Different Output Combinations. 132

(a) Number of street intersections (INT)

(b) Share of 40+ year old houses (AGE_P40) Figure 5.14 Scope Scores for Different Values of Electricity and Natural Gas Sales, and for Different Levels of Site-specific, Geographic, and Company-specific Variables: (a) Number of street intersections, (b) Share of 40+ year old houses

133

The results for the share of 40+ year old houses are very close, and similar to those presented in Table 5.10 and Figure 5.13 (where the mean age variable is used). Economies of scope are higher when the share of older houses is maximal, which suggests that older neighborhoods with paved sidewalks may provide more opportunities for efficient design of joint underground systems, hence for lower costs. The largest difference takes place when varying the number of street intersections. Higher economies of scope are achieved for the maximum number of intersections, suggesting that such intersections facilitate the joint design of both networks, thus reducing costs. The overall lowest score (0.075) is obtained for the maximal outputs and lowest number of street intersections.

5.3 Summary

Electricity and natural gas distribution systems are composed of various components, but only aggregate distribution investment costs have been analyzed in this chapter. The Box-Cox functional form turned out to be superior to the log-log and linear forms in all cases of cost estimation. The number of residential customers, residential sales and commercial-industrial sales are significant for both electricity and natural gas aggregate capital costs. As expected, these output variables are positively related to costs. The number of street intersections is the urban-related variable that is common to both models. It also has a positive coefficient, as more intersections are likely to render network operations more complicated, with the use of more capital and labor.

134

Population density has a negative effect on electricity investment costs, a result consistent with earlier research. Low-density areas with larger plots require longer conductor lines and more associated equipment than high-density areas. Higher soil corrosivity requires special materials and use of conduits to protect lines and equipment, leading to higher costs. The load factor, a proxy for capacity usage, has a negative effect on costs: the higher the load factor the lesser the capacity required per unit of annual sales. The built-up area has a positive effect on natural gas capital investment: for a given market, the larger the territory the lower the density, hence the higher the costs. Therefore, this variable is a proxy for density. Much of the natural gas system is underground, hence soil workability and the wage per employee have significant impacts on investments: the former decreases and the latter increases total capital costs. In the case of electricity, very slight economies of scale are observed in the case of market expansion at constant density. However, stronger economies are achieved with market densification. Stronger economies of scale and economies of density are observed in the case of natural gas distribution. Changes in economies of scale and density have been assessed for different levels of customer size and for different levels of the sitespecific, geographic, and company-specific variables. Lower density, higher soil corrosivity, more street intersections and lower load factors lead to higher economies of scale and higher economies of density in electricity distribution. Larger built-up areas, lower levels of soil workability, more street intersections, and higher average wages allow for more cost reductions, and also higher levels of economies of density in natural gas distribution. 135

Finally, economies of scope are evident in electricity and natural gas distribution utilities. It is more cost-efficient to serve electricity and natural gas together, than to serve electricity and natural gas only. Scope scores vary between 0.234 and 0.771 for different levels of electricity and natural gas outputs at the mean values of the site-specific variables. Scope scores decrease with increasing electricity and natural gas outputs. When different values of the site-specific variables are considered, the number of street intersections turns out to have the greatest impact. Economies of scope are lowest for the minimum number of street intersections. Scope scores are never negative, whatever the output combinations. Hence, there is no evidence of diseconomies of scope.

136

CHAPTER 6 ANALYSES OF DISAGGREGATE CAPITAL INVESTMENTS IN ELECTRICITY AND NATURAL GAS DISTRIBUTION

This chapter focuses on disaggregate-level capital investments in electricity and natural gas distribution. Electricity distribution components include: overhead conductors, underground conductors, conduits, poles, transformers, station equipment, services, and street lighting. Overhead and underground conductors are further disaggregated as primary overhead conductors, secondary overhead conductors, primary underground conductors, and secondary underground conductors. Natural gas distribution components include: mains, services, measurement and regulation stations, and industrial measurement and regulation stations. A cost function is defined for each component, and detailed elasticity and economies of scale analyses are carried out. Total distribution costs are then defined as the sums of these individual costs, and comprehensive marginal costs and economies of scale analyses are carried on these total costs. Finally, economies of scope are analyzed, with a focus on underground electricity conductors and conduits, and natural gas mains.

137

6.1 Disaggregate Cost Functions

Component capital costs are estimated in a way similar to aggregate capital cost investments, and are defined as functions of numbers of customers and sales in different sectors (residential and commercial-industrial), urban site-specific variables, geographic factors, and company-specific variables2. Log-log and Box-Cox forms3 are compared for network components, such as conductors, conduits, mains etc. The comparison is based on the log-likelihood ratio test. Mixed-form models (Box-Cox on the left-hand side and logarithms on the right-hand side) are considered when the Box-Cox parameter λ turns out to be insignificantly different from zero, leading to Equation (6.1). End-user-related components, such as services and street lighting equipment, are estimated with a linear additive form (Eq. 6.2).

(6.1)

(6.2)

Elasticities are also computed in a way similar to the elasticity calculations for the aggregate cost functions4. In the mixed-form models, the elasticities are computed by setting lambda equal to zero (Eq. 6.3). The linear additive form elasticites are computed following Equation (6.4). 2

See Section 5.1, Eq. 5.1 for a discussion of investment cost functions. See Section 5.1, Eqs. 5.2 and 5.3, for the definition of the log-log and Box-Cox forms. 4 See Section 5.1 for elasticity calculations in the case of log-log and Box-Cox forms. 3

138

(6.3)

(6.4)

Using the computed elasticities, economies of scale and density are also estimated for each component at the sample mean.

6.2 Cost Models for Electricity Distribution Network Components

The electricity distribution network is composed of conductors, services, poles, transformers and other system equipment. Some components, such as overhead and underground conductors, can be further disaggregated in terms of primary and secondary voltage, while some services can be further disaggregated in terms of overhead and underground. Conductors, poles and transformers are network components which constitute the backbone of the distribution system, while services and street lighting equipment are components closely related to customer characteristics.

6.2.1 Cost Model for Overhead Conductors

Overhead conductors (CO) are one of the backbone network components of the electricity distribution plant, together with the underground conductors. Conductors are the lines carrying power from distribution substations to the service lines. Overhead conductors cost less than underground conductors, because they do not require 139

excavation and earth filling. The number of customers, sales, and density are expected to affect costs. Soil variables, on the other hand, are not expected to have a significant effect on costs, except for topography, which may affect installation conditions. Overhead lines can be aesthetically unpleasing and may be unable to carry the high load of high-demand areas. Therefore, in wealthy neighborhoods and high-density central city areas, underground conductors may be preferred or necessary. In addition to density, the housing pattern may capture effects on costs. Low-rise detached and single-unit attached housing areas may be expected to have more overhead conductors than high-rise housing areas. Finally, an increase in the load factor, a proxy for capacity usage, is expected to decrease costs. The selected model is:

CCO = F(NRE, SRE, DENSA, SLP_P5PL, INCAGG, HOUSDETATT ,LFE) where CCO = Overhead conductor investment costs ($) NRE = Number of residential electricity customers (#) SRE = Residential electricity sales (kWh) DENSA = Area density (Population/square miles of tax district area) SLP_P5PL = Share of the land area with slopes of 5% and more (%) INCAGG = Aggregate income ($) HOUSDETATT= Number of detached and single-unit attached houses (#) LFE = Electricity load factor (Average hourly kWh sales/Hourly kWh peak load)

Descriptive statistics for the above variables are presented in Table 6.1. 140

(6.5)

Variable CO ($) NRE (#) SRE (kWh) DENSA (people/sq.m.) SLP_P5PL (%) INCAGG ($) HOUSDETATT (#) LFE

Minimum 220 9 83,415 14 0 702,810 12 0.46

Maximum 39,800,000 144,529 1,040,000,000 103,058 0.92 4,370,000,000 130,213 0.63

Mean 2,185,601 6,762 43,300,000 3,493 0.20 169,000,000 4,991 0.53

Std. Deviation 4,761,752 16,807 109,000,000 7,254 0.18 426,000,000 13,098 0.07

Table 6.1 Descriptive Statistics for the Overhead Conductor Investment Cost Model (n=241)

The log-likelihood test shows that the Box-Cox form is superior to the log-log form (Table 6.2). All the variables are significant, with the expected signs. The output variables – number of residential customers and residential sales – increase costs. Commercial-industrial sector variables turn out to be insignificant. The load factor has a negative impact, which indicates that more efficient use of capacity results in capacity savings, hence in reduced costs. The larger the shares of steeper slopes and dispersed housing, the higher the costs. The wealthier the area, the lesser the investment in overhead conductors. Elasticities are calculated at the sample mean (Table 6.3). A 1% increase in other output or dispersed housing pattern leads to about 0.4% increase in costs. A 1% increase in density reduces costs by 0.13%, while a 1% increase in steeper slopes leads to a 0.07% increase in costs. A 1% increase in income or load factor leads to decreases in costs by 0.4% and 1.4%, respectively.

141

Coefficient Constant NRE SRE DENSA SLP_P5PL INCAGG HOUSDETATT LFE

Models Log-log 5.670 (3.42)b 0.440 (2.37) 0.660 (3.17) -0.118 (-2.06) 0.001 (2.99) -0.526 (-3.10) 0.387 (2.69) -1.709 (-3.36)

λ

Box-Cox(λ, θ)a -153.2 (-5.50) 7.053 (3.52) 0.998 (4.45) -3.240 (-4.72) 21.447 (4.19) -0.567 (-4.15) 10.854 (5.67) -325.461 (-7.24) 0.259 (0.000)c 0.360 (0.000)

θ R2

0.747

0.928

Log-likelihood -3592.3239 H0: θ=λ=0d, Chi-square=260.11 H0: θ=λ=1, Chi-square=481.12 a Selected model b t-statistics in parentheses c p-value in parentheses d Log-likelihood test results

-3462.2669 (0.000) (0.000)

Table 6.2 Overhead Conductors Cost Function Estimates (n=241)

Elasticity

εNRE εSRE εDENSA εSLP_P5PL εINCAGG εHOUSDETATT εLFE

Sample Mean 0.339 0.465 -0.131 0.069 -0.376 0.483 -1.354

Table 6.3 Overhead Conductor Cost Elasticities at the Sample Mean (n=241) 142

Economies of scale and economies of density are computed at 0.804 and 0.672 at the sample mean. As expected, economies from densification are greater than economies from service expansion. Overhead conductors can be disaggregated further as primary overhead conductors and secondary overhead conductors, as discussed in the following two sections.

6.2.1.1 Cost Model for Primary Overhead Conductors

Primary overhead conductors (COP) are the lines carrying power from transmission substations to large industrial areas and to primary substations in urban areas, and from these substations to secondary voltage transformers. The wire capacity is higher in primary conductors, and carries larger loads.

COP investment costs were expected to be modeled with a cost function similar to the CO cost function (Eq. 6.5). However, the selected model (Eq. 6.6) has only 3 variables, one of which (commercial-industrial sales) does not appear to be significant at the aggregate conductor level. The difference may in part result from more limited data availability. Indeed, the sample size narrows down to 137, as only 2 companies, Central Hudson and LILCO, have primary overhead conductors data. The difference may also be structural, as primary conductors serve directly large non-residential customers.

= F(SRE, SCIE, DENSBLTP)

(6.6) 143

where = Primary overhead conductor investment costs ($) SRE = Residential electricity sales (kWh) SCIE = Commercial-industrial electricity sales (kWh) DENSBLTP = Built-up area density (Population/Total square miles of built-up area)


Variable ($) SRE (kWh) SCIE (kWh) DENSBLTP

Minimum 515 216,939 1,760 142

Maximum 26,300,000 1,040,000,000 937,000,000 18,657

Mean 1,456,618 42,600,000 48,300,000 4,386

Std. Deviation 3,843,722 132,000,000 148,000,000 4,013

Table 6.4 Descriptive Statistics for the Primary Overhead Conductor Investment Cost Model (n=137)

The estimation results are presented in Table 6.5. The Box-Cox form is superior to the log-log form. Both residential and commercial-industrial sales are significant variables. This result is plausible, because primary lines carry power to the whole market, whether serving customers directly in the case of large loads, and indirectly through transformers and secondary lines. The density variable is more precise in this model: built-up density has a higher significance than area density. The Box-Cox form is superior to the log-log form according to the log-likelihood test.

144

Coefficient Constant SRE SCIE DENSBLTP

Models Log-log Box-Cox(λ, θ)a -1.166 -17.86 (-1.01) (-1.1) 0.879 1.250 (7.95) (11.96) 0.138 0.210 (2.06) (2.49) -0.304 -2.764 (-3.63) (-5.18)

λ

0.245 (0.000) 0.330 (0.000)

θ R2

0.691

0.916

Log-likelihood -1934.3287 H0: θ=λ=0d, 115.01 H0: θ=λ=1, 356.73 a Selected model b t-statistics in parentheses c p-value in parentheses d Log-likelihood test results

-1876.825 (0.000) (0.000)

Table 6.5 Primary Overhead Conductors Cost Function Estimates (n=137)

The elasticity computations show that SRE has the greatest impact on cost increase, whereas SCIE and DENSBLTP have impacts of similar magnitudes, but with opposite signs (Table 6.6). Slight economies of scale are observed, with while economies of density are stronger, with

Elasticity

εSRE εSCIE εDENSBLTPD

=0.952,

=0.763.

Sample Mean 0.812 0.141 -0.189

Table 6.6 Primary Overhead Conductor Cost Elasticities at the Sample Mean (n=137) 145

6.2.1.2 Cost Model for Secondary Overhead Conductors

Secondary overhead conductors (COS) are the lines carrying power to residential and smaller local commercial areas. Secondary lines data is available for only 2 companies, Central Hudson and LILCO. The residential output and housing pattern are expected to be more significant in this model. The selected model (Eq. 6.7) indicates that the number of residential customers and the share of detached houses have significant effects on costs, together with built-up density and the share of steeper slopes.

= F(NRE, DENSBLTP, SLP_P5PL, HOUS_PDET)

(6.7)

where = Secondary overhead conductor investment costs ($) NRE = Number of residential electricity customers (#) DENSBLTP = Built-up area density (Population/Total square miles of built-up area) SLP_P5PL = Share of the land area with slopes 5% and more (%) HOUS_PDET= Share of detached houses (%)


146

Variable ($) NRE (#) DENSBLTP SLP_P5PL (%) HOUS_PDET (%)

Minimum 530

Maximum 13,700,000

Mean 800,250

Std. Deviation 2,216,324

33 142 0 0.03

144,529 18,657 0.72 1

6,175 4,327 0.15 0.74

2,216,324 4,005 0.17 0.22

Table 6.7 Descriptive Statistics for the Secondary Overhead Conductor Investment Cost Model (n=137)

The regression results show that all the variables are significant, with the expected signs. The Box-Cox form is superior to the log-log form (Table 6.8).

Coefficient Constant NRE DENSBLTP SLP_P6PL HOUS_PDET

Models Log-log Box-Cox(λ, θ)a 7.047 -153.2 (12.53)b (-6.09) 0.991 9.938 (19.14) (51.41) -0.234 -2.122 (-2.82) (-5.62) 0.349 35.964 ( 2.24) (3.47) 0.001 9.388 (2.72) (4.60)

λ

0.265 (0.000)c 0.337 (0.000)

θ R2

0.769

0.954


-1752.5805 (0.000) (0.000)

Table 6.8 Secondary Overhead Conductors Cost Function Estimates (n=137) 147

The elasticities at the sample mean indicate that the number of residential customers has the greatest impact on costs (Table 6.9). A 1% increase in the number of residential customers leads to a 4.3% increase in costs. The built-up density and share of steeper slopes have similar elasticities, but with opposite signs. The share of detached houses has the smallest impact (0.37).

Elasticity

εNRE εDENSBLTP εSLP_P6PL εHOUS_PDET

Sample Mean 4.288 -0.833 0.932 0.370

Table 6.9 Secondary Overhead Conductor Cost Elasticities at the Sample Mean (n=137)

Diseconomies are observed for both scale and density, with

=4.288 and

=3.455, respectively.

6.2.2 Cost Model for Underground Conductors

Underground conductors (CU) are electricity lines installed underground, whether running through conduits for durability and safety or simply buried in trenches. Underground conductor investment costs are higher than overhead conductor investment costs, due to excavation and filling expenses. Therefore, underground conductors may not be economically feasible in all areas. In fact, out of 241 districts, 16 have no underground 148

investments. Output levels, density, and community wealth and preferences are likely to determine underground investments. The selected model is:

= F(SRE, SCIE, DENSBLTP, VAL, STO_P13PL, LFE)

(6.8)

where = Underground conductor investment costs ($) SRE = Residential electricity sales (kWh) SCIE = Commercial-industrial electricity sales (kWh) DENSBLTP = Built-up area density (Population/Total square miles of built-up area) VAL = Median house value ($) STO_P13PL = Share of buildings with 13 and more stories (%) LFE = Electricity load factor (Average hourly kWh sales/Hourly kWh peak load)


Variable ($) SRE (kWh) SCIE (kWh) DENSBLTP STO_P13PL (%) VAL($) LFE

Minimum 220 9 83,415 14 0 702,810 0.46

Maximum 39,800,000 144,529 1,040,000,000 103,058 0.92 4,370,000,000 0.63

Mean 2,185,601 6,762 43,300,000 3,493 0.02 169,000,000 0.53

Std. Deviation 4,761,752 16,807 109,000,000 7,254 0.18 426,000,000 0.07

Table 6.10 Descriptive Statistics for the Underground Conductor Investment Cost Model (n=225)

149

The regression results (Table 6.11) show that the Box-Cox form is superior to the log-log form. However, λ turns out to be insignificant, and, therefore, a mixed model is estimated where the Box-Cox transformation is only applied to the cost variable and the logarithm is applied to the explanatory variables. The mixed model is still superior to the log-log one, with all variables being significant and having the expected signs.

Coefficient Constant SRE SCIE DENSBLTP VAL STO_P13PL LFE

Models Mixed Box-Cox Model -6.746 -80.59 (-5.04)a (-12.30) 0.821 4.327 (6.30) (6.79) 0.246 1.078 (6.30) (3.45) -0.250 -0.904 (-3.02) (-2.23) 0.230 2.620 (1.82) (4.23) 0.001 0.006 (3.04) (4.96) -2.195 -11.332 (-3.51) (-3.71)

Log-log

λ θ R2

0.000 0.139 (0.000)b 0.757

0.817

Log-likelihood -3089.245 H0: θ=λ=0c, Chi-square = 55.14 H0: θ=λ=1, Chi-square = 1116.52 a t-statistics in parentheses b p-value in parentheses c Log-likelihood test results

-3061.6741 (0.000) (0.000)

Table 6.11 Underground Conductors Cost Function Estimates (n=225)

150

Residential and commercial-industrial electricity sales increase the costs. Built-up density and share of high-rise buildings are also significant, but with opposite signs. Higher densities decrease investment costs, which can be explained by the use of less and shorter lines. However, increasing the share of high-rise buildings leads to higher costs, which may be explained by congestion-related expenses. The median house value, used as a proxy to the wealth of the area, has a positive effect on costs, which may be explained by more demand for underground conductors due to aesthetic concerns and willingness to pay for these expenses. Better use of capacity, as measured by an increasing LFE, decreases costs, as expected. The elasticity analysis shows that LFE has the highest impact on costs (Table 6.12). A 1% increase in LFE decreases costs by 8.5%. A 1% increase in SRE increases costs three times more than a 1% increase in SCIE. When VAL goes up by one percent, costs increase by almost 2%. DENSBLTP and STO_P13PL are comparatively less influential. Diseconomies of scale and density are observed at the sample mean, with =4.053 and

Elasticity

εSRE εSCIE εDENSBLTP εVAL εSTO_P13PL εLFE

=3.374, respectively.

Sample Mean 3.245 0.808 -0.678 1.965 0.005 -8.499

Table 6.12 Underground Conductor Cost Elasticities at the Sample Mean (n=225)

151

Underground conductors can be disaggregated into primary and secondary underground conductors, as discussed in the next two sections.

6.2.2.1 Cost Model for Primary Underground Conductors

Primary underground lines (CUP) carry power to urban built-up areas and large industries, in a way similar to primary overhead lines. The sample size is smaller than in the case of aggregate underground conductors, because disaggregate data is available for only two companies: Central Hudson and LILCO. The cost model is expected to be similar to the aggregate underground conductor cost model, while capturing site-specific characteristics peculiar to primary lines. The selected model (Eq. 6.9) has an additional old housing share variable, and the median household income is used as a proxy to the wealth of the area instead of the median house value. The share of old housing stock captures the effect of the concentration of old neighborhoods, which are generally located in the central city, and may require special pavement. The selected model is:

= F(SRE, SCIE, DENSBLTP, INC, STO_P13PL, LFE, AGE_P40PL)

(6.9)

where = Primary underground conductor investment costs ($) SRE = Residential electricity sales (kWh) SCIE = Commercial-industrial electricity sales (kWh) DENSBLTP = Built-up area density (Population/Total square miles of built-up area) INC = Median household income ($) 152

STO_P13PL = Share of buildings with 13 and more stories (%) LFE = Electricity load factor (Average hourly kWh sales/Hourly kWh peak load) AGE_P40PL = Share of houses 40 years and older (%)

Descriptive statistics for the above variables in are presented in Table 6.13.

Variable

($) SRE (kWh) SCIE (kWh) DENSBLTP INC ($) STO_P13PL (%) LFE AGE_P40PL (%)

Minimum

Maximum

Mean

15 1,041,822 1,760 343 13841 0.000 0.46 0.03

14,900,000 1,040,000,000 937,000,000 18,657 2580752 0.156 0.60 0.90

848,130 46,700,000 53,100,000 4,551 139956 0.004 0.48 0.35

Std. Deviation 2,243,977 139,000,000 155,000,000 4,114 366317 0.018 0.05 0.18

Table 6.13 Descriptive Statistics for the Primary Underground Conductor Investment Cost Model (n=124)

The regression results point to the superiority of the Box-Cox form over log-log form. However, the Box-Cox parameter λ turns out to be insignificant, and a mixed model is estimated where the Box-Cox transformation is applied to the cost, and the logarithm to the explanatory variables.

153

Coefficient Constant SRE SCIE DENSBLTP INC STO_P13PL LFE AGE_P40PL

Log-log -4.659 (-2.36) a 0.662 (2.70) 0.381 (3.76) -0.542 (-4.68) 0.284 (1.63) 0.001 (3.19) -2.905 (-2.29) 0.424 (2.27)

Models Mixed Box-Cox Model -54.065 (-3.03) 3.234 (3.38) 1.301 (3.29) -1.987 (-4.40) 2.202 (2.65) 0.006 (4.13) -9.688 (-2.00) 1.493 (2.05)

λ θ R2

0.000 0.127 (0.000) b 0.728

0.797

Log-likelihood -1619.4535 H0: θ=λ=0c, Chi-square = 25.15 H0: θ=λ=1, Chi-square = 618.93 a t-statistics in parentheses b p-value in parentheses c Log-likelihood test results

-1606.88 (0.000) (0.000)

Table 6.14 Primary Underground Conductors Cost Function Estimates (n=124)

The elasticities at the sample mean (Table 6.15) indicate that a 1% increase in SRE would increase costs by 1%, while a 1% increase in SCIE would increase costs by 0.4%. The load factor has the greatest impact, with a 1% increase in LFE leading to a 3% decrease in costs. DENSBLTP and INC have similar impacts on costs (0.6%), but in opposite ways. The share of high-rise buildings, STO_P13PL, has the smallest impact on costs. A 1% increase in the share of older houses decreases costs by 0.47%. 154

Diseconomies of scale take place, with through densification, with

Elasticity

εSRE εSCIE εDENSBLTP εINC εSTO_P13PL εLFE εAGE_P40PL

= 1.415. However, economies are achieved

0.795.

Sample Mean 1.009 0.406 -0.620 0.687 0.002 -3.023 0.466

Table 6.15 Primary Underground Conductor Cost Elasticities at the Sample Mean (n=124)

6.2.2.2 Cost Model for Secondary Underground Conductors

Secondary underground lines (CUS) carry power to residential and small commercial areas. The variables for this cost model are the same as those for primary underground conductors, except for the median household income variable. Also the residential density has a higher significance than the built-up density. The selected model is:

= F(SRE, SCIE, DENSRESID, STO_P13PL, LFE, AGE_P40PL) where = Secondary underground conductor investment costs ($) SRE = Residential electricity sales (kWh) 155

(6.10)

SCIE = Commercial-industrial electricity sales (kWh) DENSRESID = Residential area density (Population/Total square miles of residential area) STO_P13PL = Share of buildings with 13 and more stories (%) LFE = Electricity load factor (Average hourly kWh sales/Hourly kWh peak load) AGE_P40PL = Share of houses 40 years and older (%)


Variable ($) SRE (kWh) SCIE (kWh) DENSRESID STO_P13PL (%) LFE AGE_P40PL ($)

Minimum 21

Maximum 4,774,939

Mean 357,498

Std. Deviation 820,037

216,939 1,760 167 0.000 0.46 0.03

1,040,000,000 937,000,000 36,453 0.156 0.60 0.90

46,800,000 53,200,000 6,241 0.004 0.48 0.35

139,000,000 156,000,000 6,293 0.018 0.04 0.18

Table 6.16 Descriptive Statistics for the Secondary Underground Conductor Investment Cost Model (n=123)

The Box-Cox form turns out to be superior to the log-log form (Table 6.17). All the variables are significant and with the expected signs. The load factor, LFE, has the greatest impact on costs according to the elasticity analysis: a 1% increase in LFE leads to a cost decrease of 3.5% (Table 6.18). A 1% increase in SRE increases costs by 0.7%. The elasticities of all the other variables are less than 0.4, and only density has a negative effect on costs. Economies of scale and density are obtained with this model, with

= 0.897 and 156

0.740.

Coefficient Constant SRE SCIE DENSRESID STO_P13PL LFE AGE_P40PL

Models Log-log -7.580 (-3.83)b 0.849 (5.43) 0.213 (2.31) -0.216 (-2.02) 0.001 (2.64) -6.316 (-4.47) 0.470 (2.59)

Box-Cox(λ, θ)a -15.7 (-1.97) 0.803 (5.81) 0.251 ( 2.51) -0.525 (-2.31) 2.636 ( 4.27) -35.817 (-4.33) 3.679 (3.07)

λ

0.117 (0.021)c 0.167 (0.000)

θ R2

0.671

0.775

Log-likelihood -1551.4848 H0: θ=λ=0d, Chi-square = 34.1 H0: θ=λ=1, Chi-square = 475.5 a Selected model b t-statistics in parentheses c p-value in parentheses d Log-likelihood test results

-1534.4334 (0.000) (0.000)

Table 6.17 Secondary Underground Conductors Cost Function Estimates (n=123)

Elasticity

εSRE εSCE εDENSRESID εSTO_P13PL εLFE εAGE_P40PL

Sample Mean 0.680 0.216 -0.156 0.148 -3.505 0.347

Table 6.18 Secondary Underground Conductor Cost Elasticities at the Sample Mean (n=123)

157

6.2.3 Cost Model for Conduits

Some underground conductors are inserted into conduits to increase safety and protect the lines from corrosion. Different materials are used for conduits, including metal, clay, and plastic. It is expected that investment in conductors have a direct effect on conduit costs. Two alternative models are considered for conduit investment costs. The first model considers the underground conductor cost at the aggregate level as the explanatory variable. In this case, the sample size (n=434) is larger, since data is available for all the tax districts of the four companies. The second model considers both primary and secondary underground conductors as the explanatory variables. The sample size (n=122) in this case is much smaller, due to the limited availability of data for only two companies.

Cost Model Alternative 1

This model considers conduit investment costs as a function of underground conductor investments, and site-specific and geographic variables. The selected model is:

= F(CCU, DENSA, INCAGG, STO_P13PL, AGE_P40PL) where = Conduit investment costs ($) = Underground conductor investment costs ($) 158

(6.11)

DENSA = Area density (Population/Total square miles of tax district area) INCAGG = Aggregate household income ($) STO_P13PL = Share of buildings with 13 and more stories (%) AGE_P40PL = Share of houses 40 years and older (%)


Variable

($) ($) DENSA (people/sq.m.) INCAGG ($) STO_P13PL (%) AGE_P40PL (%)

Minimum 20 36 2 1,781,925 1.E-08 0.01

Maximum 126,000,000 33,500,000 103,058 4,370,000,000 0.0365 0.90

Mean 779,669 729,010 2,305 109,000,000 0.0004 0.40

Std. Deviation 6,347,999 2,560,344 5,657 326,000,000 0.0027 0.20

Table 6.19 Descriptive Statistics for the Conduit Investment Cost Model Alternative 1 (n=434)

The Box-Cox form is superior to the log-log one, as in the case of most network component cost models (Table 6.20). All the variables are significant. The area density has a positive effect on costs, in contrast to density effects on conductor costs. This is most likely due to the difficulty of excavation and filling in denser areas. High-income areas tend to invest more for the safety provided by conduits. Higher concentrations of older neighborhoods and high-rise buildings also increase costs, in a way similar to underground conductor costs, since these areas may have special conditions, such as narrow roads and difficult pavement requirements. 159

Coefficient Constant

DENSA INCAGG STO_P13PL AGE_P40PL

Models Log-log -5.856 (-5.05) b 0.816 (15.69) 0.195 (5.17) 0.361 (4.27) 0.000 (1.54) 0.336 (3.18)

λ

Box-Cox(λ, θ)a 5.014 ( 2.01) 0.393 (17.99) 0.199 (5.76) 0.032 (2.49) 0.799 (2.09) 1.484 (4.11) 0.162 (0.000) c 0.100 (0.000)

θ R2

0.724

0.790


-5201.2393 (0.000) (0.000)

Table 6.20 Conduit Cost Function Estimates, Model Alternative 1 (n=434)

The elasticity analysis (Table 6.21) indicates that underground conductor investments have the largest impact on costs: a 1% increase in this variable increases conduit costs by almost 1%. An older housing stock has the second greatest impact: a 1% increase in this variable increases costs by 0.3%. All the other variables have much smaller elasticities.

160

Elasticity

εCU εDENSA εINCAGG εSTO_P13PL εAGE_P40PL

Sample Mean 0.969 0.193 0.180 0.062 0.353

Table 6.21 Conduit Cost Elasticities at the Sample Mean, Model Alternative 1 (n=434)

Cost Model Alternative 2

The second model alternative considers both primary and secondary underground conductors. The sample is limited to Central Hudson and LILCO, with 122 observations. The selected model has also density and soil-related variables (Eq.6.12).

= F(

,

, DENSA, WATDEPTH)

where = Conduit investment costs ($) = Primary underground conductor investment costs ($) = Secondary underground conductor investment costs ($) DENSA = Area density (Population/Total square miles of tax district area) WATDEPTH = Water table depth (feet)


161

(6.12)

Variable ($) ($) ($) DENSA (pop./sq.m.) WATDEPTH (feet)

Minimum 446 15

Maximum 11,200,000 14,900,000

Mean 582,398 658,205

Std. Deviation 1,588,079 1,954,262

10

4,774,939

272,803

715,630

12 1.76

18,656 5.56

3,038 3.76

3,898 0.75

Table 6.22 Descriptive Statistics for the Conduit Investment Cost Model Alternative 2 (n=122)

The Box-Cox form is again superior to the log-log one (Table 6.23). The area density is also significant, with a positive sign, as discussed in the previous model alternative. A soil-related variable, the water table depth, also turns out to be significant, with the expected negative sign: the closer the water table is to the topsoil, the higher the potential corrosivity, hence the higher the drainage costs. A 1% increase in primary conductor investments increases conduit costs by close to 7%, twice more than a 1% increase in secondary conductor cost investments (Table 6.24). A 1% increase in the depth of the water table, on the other hand, decreases costs by 0.6%. The area density has the smallest effect (0.19), exactly the same as in cost model alternative 1.

162


DENSA WATDEPTH

Models Log-log -0.096 (-0.15) b

Box-Cox(λ, θ)a 3.609 (1.80)

0.663 (11.87)

0.448 (11.16)

0.258 (4.65) 0.248 (5.21) -0.645 (-1.92)

0.256 (5.21) 0.324 (5.33) -3.406 (-2.94)

λ

0.175 (0.000) c 0.145 (0.000)

θ R2

0.820

0.893


-2052.196 (0.000) (0.000)

Table 6.23 Conduit Cost Function Estimates, Model Alternative 2 (n=122)

Elasticity

εCUP εCUS εDENSA εWATDEPTH

Sample Mean 0.687 0.336 0.194 -0.630

Table 6.24 Conduit Cost Elasticities at the Sample Mean, Model Alternative 2 (n=122)

163

6.2.4 Cost Model for Services

Services are the lines connecting primary or secondary conductors to the endusers. Service lines are considered as an end-user related component rather than a network component. They are directly related to the number of customers, since a service line is connected to each customer. Customer size, the amount of electricity sales per customer, is also expected to have an impact on service costs, particularly to capture the effect of servicing multi-unit structures. Site-specific and geographic conditions are also considered in service lines investments. Because of the previous considerations, the selected model has the form:

= NRE*f(ZRE, DENSBLTP,SOILSTEEL) + NCIE*g(ZCIE, DENSBLTP,SOILSTEEL) (6.13) where = Service lines investment costs ($) NRE = Number of residential electricity customers (#) ZRE = Residential electricity customer size (Residential kWh sales/Number of residential customers) NCIE = Number of commercial-industrial electricity customers (#) ZCIE = Commercial-industrial electricity customer size (Commercial-industrial kWh sales/Number of commercial-industrial) DENSBLTP = Built-up area density (Population/Total square miles of built-up area) SOILSTEEL = Percentage of soil susceptible to steel corrosion (%) 164


Variable ($) NRE (#) ZRE (kwh/#) NCIE (#) ZCIE (kwh/#) DENSBLTP SOILSTEEL (%)

Minimum 31

Maximum 8,312,341

Mean 117,032

Std. Deviation 471,355

8 3,351 3 880 69 0.001

144,529 25,689 17,084 1,518,895 742,621 98.99

6,706 7,561 759 88,082 36,606 47.57

471,355 3,847 1,782 138,125 545,860 24.82

Table 6.25 Descriptive Statistics for the Service Investment Cost Model (n=137)

The linear additive form is used for both functions f and g. However, none of the variables in g turned out to be significant and the variable NCIE is retained as a standalone variable. This specification was also used by Guldmann (1985) to estimate electricity service investment costs. The results are presented in Table 6.26. The residential customer size variable has non-linear effects, hence the use of

. The built-

up density has a negative effect on costs, which can be explained by the need for shorter service lines on smaller plots, or shorter distances between plot line and building line. Increased steel corrosion increases service line costs, particularly those buried underground.

165

Model Additive

Coefficient NRE*ZRE

0.008 (3.82) a -8.55E-07 (-3.75) -0.001 (-3.57) 0.302 (5.24) 295.5 (7.08)

NRE*ZRE2 NRE*DENSBLTP NRE*SOILSTEEL NCIE

a

R2 t-statistics in parentheses

0.929

Table 6.26 Service Cost Function Estimates (n=137)

In order to compute correct elasticities (Eqs. 6.15-6.19), the service cost function in Table 6.26 must be reformulated by replacing ZRE by the ratio SRE/NRE (Eq. 6.14):

CSERVE = 0.008 SRE – 8.55*10-7SRE2/NRE – 0.001NRE*DENSBLTP + 0.302NRE*SOILSTEEL + 295.5NCIE

(6.14)

0.008 – 17*10-7

(6.15)

-8.55*10-7*(-1)

– 0.001 DENSBLTP + 0.302 SOILSTEEL

and (6.16)

= 295.5

and

= -0.001

(6.17)

(6.18)

and

166

= 0.302

(6.19)

and

The elasticity analysis shows that is a 1% increase in the number of residential customers increases costs by 2.5%, whereas a 1% increase in the number of commercialindustrial customers increases costs by 3.2%. Sales have a negative elasticity: a 1% increase in sales decreases service costs by 1.8%. Density has also a negative effect on costs: a 1% increase in density decreases costs by almost 3.5%. Steel corrosivity has the smallest impact on costs (1.37%).

Elasticity

εNRE εSRE εNCIE εDENS εSOILSTEEL

Sample Mean 2.538 -1.792 3.2 -3.487 1.368

Table 6.27 Service Cost Elasticities at the Sample Mean (n=137)

6.2.5 Cost Model for Poles

Poles support electricity overhead lines, transformers and street lights. They are made up of wood or metal. Overhead primary and secondary conductor investments affect the cost of pole investments directly. The pole cost function also considers density and the average wage per employee. The selected model is: 167

= F(

,

, DENSBLTP, WPEE)

(6.20)

where = Pole investment costs ($) = Primary overhead conductor investment costs ($) = Secondary overhead conductor investment costs ($) DENSBLTP = Built-up area density (Population/Total square miles of built-up area) WPEE = Wage per electricity employee ($)


Variable ($) ($) ($) DENSBLTP WPEE ($)

Minimum 11,068 1,269 1,712 160 16,622

Maximum 31,400,000 26,300,000 13,700,000 18,657 17,753

Mean 1,880,879 1,456,865 800,329 4,413 17,547

Std. Deviation 4,991,741 3,843,628 2,216,296 4,018 438

Table 6.28 Descriptive Statistics for the Pole Investment Cost Model (n=137)

The regression results (Table 6.29) show that the Box-Cox form is superior to the log-log form. All variables are significant, with the expected signs. The negative sign of density can be explained by the need for less poles in high-density urban areas. The average wage per employee has a positive sign since poles may require frequent maintenance. 168


DENSBLTP WPEE

Models Log-log Box-Cox(λ, θ)a -23.663 -294.2 (-3.21) b (-3.96) 0.372 0.365 (7.85) ( 9.02) 0.568 0.720 (11.55) (15.07) -0.044 -0.174 (-2.38) (-2.94) 2.604 5.973 (3.45) (4.15)

λ

0.280 (0.000) c 0.275 (0.000)

θ R2

0.978

0.991


-1750.4312 (0.000) (0.000)

Table 6.29 Pole Cost Function Estimates (n=137)

The elasticity results (Table 6.30) shows that secondary lines have twice the effect on cost that primary lines have (0.612 vs. 0.367). The average wage per employee, WPEE, has the greatest impact on costs: a 1% increase in WPEE leads to an increase in costs of 1.7%. DENSBLTP has a small impact (-0.034).

169

Elasticity

εCOP εCOS εDENSBLTP εWPEE

Sample Mean 0.367 0.612 -0.034 1.741

Table 6.30 Pole Cost Elasticities at the Sample Mean (n=137)

6.2.6 Cost Model for Transformers

Transformers are the local stations where electricity voltage is reduced from primary to secondary level. Energy losses are minimized when power is transmitted at high voltages. However, high voltages must be reduced for the safe use by end-users. Transformers can be attached to poles or located at ground level. Overhead and underground lines are connected to transformers for electricity voltage regulation. Overhead and underground lines, electricity sales, load factor, and geographic variables are expected to have an effect on transformer investment costs. The selected model is:

= F(CCO, CCU, SRE, SCIE, SOILCORR, SOILWORK, LFE) where = Overhead conductor investment costs ($) = Underground conductor investment costs ($) SRE = Residential electricity sales (kWh) SCIE = Commercial-industrial electricity sales (kWh) SOILCORR = Share of soil susceptible to corrosion (%) 170

(6.21)

SOILWORK = Share of workable soil (%) LFE = Electricity load factor (Average hourly kWh sales/Hourly kWh peak load)


Variable ($)

($) ($) SRE (%) SCIE (%) SOILCORR (%) SOILWORK (%) LFE

Minimum 5,035 1,045 36 216,939 1,760 2.78 1.57 0.46

Maximum 25,400,000 39,800,000 33,500,000 1,040,000,000 2,970,000,000 99.80 90.63 0.63

Mean 1,405,287 2,343,922 1,300,603 46,600,000 97,200,000 77.86 56.69 0.53

Std. Deviation 3,277,604 4,926,235 3,466,999 113,000,000 290,000,000 28.28 20.23 0.07

Table 6.31 Descriptive Statistics for the Transformer Investment Cost Model (n=225)

The Box-Cox form is superior to the log-log form (Table 6.32). All the variables are significant, with the expected signs. More conductors, both overhead and underground, increase transformer investments. Soil corrosion and workability are related to ground-level transformers. The load factor has a negative effect on costs. The elasticity results (Table 6.33) show that the load factor has the greatest impact on transformer investment costs. A 1% increase in LFE decreases costs by 0.5 %. increases transformer costs about twice more than

, and SRE increases these costs

almost six times more than SCIE. SOILCORR and SOILWORK have similar impacts on costs, but with opposite signs. 171

Models Log-log -2.163 (-4.11) 0.146 (4.61) 0.107 (4.28) 0.581 (10.45) 0.114 (4.57) 0.295 (2.65) -0.344 (-2.70) -0.799 (-2.54)

Coefficient Constant CO CU SRE SCIE SOILCORR SOILWORK LFE

Box-Cox(λ, θ)a -16.06 (-1.92) 0.231 (9.58) 0.133 (6.91) 0.100 (7.94) 0.014 (2.75) 4.032 (2.72) -4.735 (-2.61) -50.230 (-2.81)

λ

0.322 (0.000) 0.302 (0.000)

θ R2

0.925

0.965


-2998.8061 (0.000) (0.000)

Table 6.32 Transformer Cost Function Estimates (n=225)

Elasticity

εCCO εCCU εSRE εSCIE εSOILCORR εSOILWORK εLFE

Sample Mean 0.358 0.170 0.405 0.070 0.227 -0.241 -0.568

Table 6.33 Pole Cost Elasticities at the Sample Mean (n=225)

172

6.2.7 Cost Model for Street Lighting

Street lighting is not a network component, and is similar to services. Lighting customers and sales are considered as the primary variables that affect street lighting costs. Street lights are located along streets, and therefore street length is likely to affect these costs. The need for lighting is also related to urban form, and therefore density may be considered as another factor affecting street lighting costs. The selected model is:

= NL*f(ZL, STR, DENSBLTP)

(6.22)

where = Street lighting investment costs ($) NL = Number of lighting customers (#) ZL = Lighting customer size (Lighting kWh sales/Number of lighting customers: S L/NL) STR = Total street length (miles) DENSBLTP = Built-up area density (Population/Total square miles of built-up area)


Variable CSTL ($) NL (#) ZL (kwh/#) STR (miles) DENSBLTP

Minimum 981 4 780 0.19 142

Maximum 44,186,279 630 7,803,984 2,363 220,188

Mean 1,054,568 25 180,117 116 5,200

Std. Deviation 3,710,801 61 661,435 238 14,478

Table 6.34 Descriptive Statistics for the Street Lighting Investment Cost Model (n=233) 173

A linear additive form is selected for the function f in Eq. (6.22). This form was also used by Guldmann (1985, 1988) in street lighting cost modeling. All the variables are significant, with the expected signs.

Coefficient NL*ZL NL*STR NL*DENSBLTP R2

Model Additive 0.828 (32.55) 10.848 (3.01) -0.656 (-3.17) 0.879

Table 6.35 Street Lighting Cost Function Estimates (n=233)

As for the services cost function, the street lighting cost function must be reformulated (Eq. 6.23) in terms of sales and number of customers in order to compute elasticities (Eqs. 6.24-6.27):

(6.23)

(6.24)

(6.25)

174

(6.26)

(6.27)

The elasticity analysis (Table 6.36) show that the number of lighting customers and street length have almost the same effect on costs. A 1% increase in these variables leads to a 0.89% increase in costs. Lighting sales have a smaller impact (0.043). A 1% increase in built-up density decreases costs by 0.002%.

Elasticity

εNL εSL εSTR εDENSBLTP

Sample Mean 0.894 0.043 0.896 -0.002

Table 6.36 Street Lighting Cost Elasticities at the Sample Mean (n=233)

6.2.8 Cost Model for Station Equipment

Electricity stations step down the voltage from sub-transmission levels to primary levels. Electricity sales are the main factor affecting the size and number of stations. Capacity utilization, as provided by the load factor, is also an influential variable. The selected model is:

175

= F(SRE, SCIE, LFE)

(6.28)

where SRE = Residential electricity sales (kWh) SCIE = Commercial-industrial electricity sales (kWh) LFE = Electricity load factor (Average hourly kWh sales/Hourly kWh peak load)


Variable STEQ ($) SRE (kWh) SCIE (kWh) LFE

Minimum 3,000 1,282,618 495,542 0.46

Maximum 39,300,000 1,040,000,000 2,970,000,000 0.63

Mean 2,859,309 72,600,000 158,000,000 0.56

Std. Deviation 5,337,267 141,000,000 366,000,000 0.07

Table 6.37 Descriptive Statistics for the Station Equipment Investment Cost Model (n=133)

The Box-Cox form is superior to the log-log form (Table 6.38). Sales in both the residential and commercial-industrial sectors, and the load factor turn out to be statistically significant, with the expected signs. The elasticity analysis (Table 6.39) shows that LFE has the greatest impact on costs. A 1% increase in SRE leads to a cost increase of 0.65%, while a 1% increase in SCIE leads to a cost increase of 0.26%. Slight economies of scale are observed at the sample mean, with

= 0.912.

176

Coefficient Constant SRE SCIE LFE

Models Log-log -1.290 (-1.05) b 0.515 (3.65) 0.308 (2.86) -1.526 (-1.98)

Box-Cox(λ, θ)a -8.326 (-0.24) 0.093 (7.48) 0.028 (3.40) -149.542 (-2.36)

λ

0.368 (0.000) c 0.319 (0.000)

θ R2

0.586

0.777


-1998.9 (0.000) (0.000)

Table 6.38 Station Equipment Cost Function Estimates (n=133)

Elasticity

εSRE εSCE εLFE

Sample Mean 0.651 0.261 -1.081

Table 6.39 Station Equipment Cost Elasticities at the Sample Mean (n=233)

177

6.3 Cost Models for Natural Gas Distribution Network Components

The natural gas distribution network is made up of mains, services and other system equipment. The disaggregate level cost analysis is limited to 4 components, due to the unavailability of more detailed data. Distribution mains are the backbone of the network, while services, measurement and regulation stations and industrial measurement and regulation stations are the components directly related to end-user characteristics.

6.3.1 Cost Model for Distribution Mains

Distribution mains are the fundamental component of the natural gas network. Mains are made of steel, iron and plastic, and are buried underground. Gas sales in the residential and commercial-industrial sectors, site-specific, geographic factors, and company-specific factors are expected to affect costs. The selected model is:

= F(SRG, SCIG, HOUSDET, WATDEPTH, WPEG, HDDAVG) where SRG = Residential gas sales (mcf) SCIG = Commercial-industrial gas sales (mcf) HOUSDET= Number of detached houses (#) WATDEPTH = Water table depth (feet) WPEG = Wage per gas employee ($) 178

(6.29)

HDDAVG= Average heating degree days (°F) Descriptive statistics for the above variables are presented in Table 6.40.

Variable ($) SRG (mcf) SCIG (mcf) HOUSDET (#) WATDEPTH (feet) WPEG ($) HDDAVG (°F)

Minimum 33,455 8 4 10 0.80 15,336 454

Maximum 89,000,000 7,493,316 5,541,700 128,723 5.74 20,550 677

Mean 6,130,456 411,433 311,081 5,309 3.61 18,144 517

Std. Deviation 11,500,000 864,589 670,081 13,887 0.86 2,387 52

Table 6.40 Descriptive Statistics for the Distribution Mains Investment Cost Model (n=190)

The Box-Cox and log-log forms are compared, using the log-likelihood test, and the Box-Cox form turns out to be superior (Table 6.41). Increasing natural gas sales leads to increases in costs. Natural gas is primarily used for space heating. The number of detached single houses turns out to be significant, and may be related to their increasing need of natural gas for space heating. The average heating degree-days are also significant for the same reason. Colder areas, where the average heating degree-days are high, require more investment in the capacity of the natural gas network to be able to serve customers at the peak time. The average wage per employee is also expected to impact costs, as underground excavation, installation and filling are labor intensive. A high water table necessitates drainage, and makes excavation and filling more difficult. Thus, the shallower the water table, the higher the costs. 179

Coefficient Constant SRG SCIG HOUSDET WATDEPTH WPEG HDDAVG

Models Log-log -19.343 (-3.69) b 0.421 (15.28) 0.104 (4.60) 0.401 (12.15) -0.299 (-2.32) 1.302 (4.13) 1.998 (4.39)

λ

0.146 (0.000) c 0.186 (0.000)

θ R2

Box-Cox(λ, θ)a -232.102 (-5.52) 1.332 (15.32) 0.242 (3.21) 1.955 (11.87) -3.808 (-2.52) 4.841 (4.68) 13.088 (5.35)

0.929

0.947


-2838.5821 (0.000) (0.000)

Table 6.41 Distribution Mains Cost Function Estimates (n=190)

The cost elasticities are calculated at the sample mean (Table 6.42). The impact of SRG (0.469) is almost six times the impact of SCIG (0.082). HDDAVG and WPEG have higher impact on costs: a 1% increase in these variables leads to increases in costs by 1.7% and 1%, respectively. HOUSDET and WATDEPTH have smaller elasticities, but with opposite signs. Economies of scale are observed for distribution mains, with

=

0.550. However, such economies would take place with a fixed number of detached houses, which may be viewed as keeping the service territory fixed. Hence, 180

=

0.550 should be viewed as a measure of economies of density. If the territory, i.e. the number of detached houses, expands, then the economies of scale are 0.550+0.364=0.914.

Elasticity

εSRG εSCIG εHOUSDET εWATDEPTH εWPEG εHDDAVG

Sample Mean 0.469 0.082 0.364 -0.244 1.079 1.732

Table 6.42 Distribution Mains Cost Elasticities at the Sample Mean (n=190)

6.3.2 Cost Model for Gas Services

Service pipes carry natural gas from mains to end-users. Service pipes are buried underground, in a way similar to distribution mains. Service costs are expected to be related to customers and sales, as well as site-specific and soil characteristics. The selected model is:

= NRG f(ZRG, DENSBLTP, HOUSDET, SOILSTEEL, WATDEPTH) + NCIG g(ZCIG, DENSBLTP, HOUSDET, SOILSTEEL, WATDEPTH) where NRG = Number of residential gas customers (#)

181

(6.30)

ZRG = Residential gas customer size (Residential gas sales (mcf)/Number of residential customers (#)) NCIG = Number of commercial-industrial gas customers (#) ZCIG = Commercial-industrial gas customer size (Commercial-industrial gas sales (mcf)/Number of commercial-industrial customers (#)) DENSBLTP = Built-up area density (Population/Total square miles of built-up area) HOUSDET= The number of detached houses (#) SOILSTEEL = Percentage of soil susceptible to steel corrosion (%) WATDEPTH = Water table depth (feet)


Variable SERVG ($) NRG (#) ZRG (mcf/#) NCIG (#) ZCIG (mcf/#) DENSBLTP (people/sq m.) HOUSDET (#) SOILSTEEL (%) WATDETPH (feet)

Minimum 111 1 3 1 4 401 10 0.01 0.80

Maximum 25,049,664 71,583 279 5,502 16,128 220,188 129,835 99.98 5.74

Mean 1,166,955 4,326 107 365 1,045 5,575 5,713 27.79 3.61

Std. Deviation 2,687,762 8,425 59 737 1,824 15,753 16,186 23.04 0.86


A linear additive model is selected for the functions f and g in Eq. (6.31), in a way similar to the electricity service cost model (Table 6.44). In the case of residential 182

customers, the average cost per customer increases with customer size and the number of detached houses. It decreases with an increasing depth of the water table. The relationship between service cost and density is non-linear, hence the use of the squared density. Detached houses are single units located on larger plots, which require longer service pipes, hence increased costs. Finally, steel corrosivity affects service costs. In the case of commercial-industrial customers, all variables, except ZCIG, turned out insignificant. Hence, the commercial-industrial part of the service cost model is reduced to the variable SCIG.

Coefficient NRG*ZRG SCIG NRG*DENSBUILTP NRG*DENSBUILTUP2 NRG*HOUSDET NRG*SOILSTEEL NRG*WATDEPTH R2

Model Additive 2.010 (8.91) 0.609 (6.55) 0.007 (3.57) -3.5E-08 (-3.4) 0.0004 (3.97) 1.096 (4.90) -8.437 (-2.60) 0.977

Table 6.44 Service Cost Function Estimates (n=190)

The cost function is reformulated to compute correct elasticities (Eqs. 6.32-6.38):

183

(6.31)

(6.32)

(6.33)

(6.34)

(6.35)

(6.36)

(6.37)

(6.38)

The elasticity analysis shows that residential sales have the greatest impact on costs: a 1% increase in this variable leads to a 62% increase in costs. The number of residential customers, and commercial-industrial sales have very close elasticities (0.14). A 1% increase in built-up density increases costs by 0.128%. Steel corrosivity and the 184

depth of water table have impacts of similar magnitudes, but with opposite signs. The number of detached houses has less impact on costs: a 1% increase in this variable leads to a 0.008% increase in costs.

Elasticity

εNRE εSRE εSCIG εDENSBLTP εHOUSDET εSOILSTEEL εWATDEPTH

Sample Mean 0.139 0.616 0.140 0.128 0.008 0.102 -0.102

Table 6.45 Service Cost Elasticities at the Sample Mean (n=190)

6.3.3 Cost Model for Measurement and Regulation Stations

Measurement and regulation stations regulate gas pressure for residential and small commercial customers. These stations are located above ground and are closely related to customer characteristics. The selected model follows the structure of the service cost model:

= NRG f(ZRG, HDDAVG, ROCKDEPTH) + NCIG g(ZCIG, HDDAVG, ROCKDEPTH) (6.39) where NRG = Number of residential gas customers (#)

185

ZRG = Residential gas customer size (Residential gas sales (mcf)/Number of residential customers (#)) NCIG = Number of commercial-industrial gas customers (#) ZCIG = Commercial-industrial gas customer size (Commercial-industrial gas sales (mcf)/Number of commercial-industrial customers (#)) HDDAVG = Average heating degree days (°F) ROCKDEPTH = Average rock depth (inches) Descriptive statistics for the above variables are presented in Table 6.46.

Variable CMRS ($) NRG (#) ZRG (mcf/#) NCIG (#) ZCIG (mcf/#) ROCKDEPTH (inches) HDDAVG (°F)

Minimum 205 4 6 1 23 19.31 454

Maximum 3,436,000 71,583 264 5,502 16,128 60.00 677

Mean 180,839 6,039 104 516 1,272 51.46 561

Std. Deviation 328,419 9,786 54 863 2,250 6.29 63


Data is available for three companies: Central Hudson, LILCO, and Niagara Mohawk. Therefore, the sample size decreases to 135. In the case of commercialindustrial customers, all the variables, except ZCIG, turned out insignificant. Hence, the commercial-industrial part of the model is reduced to the variable SCIG. A linear additive form is used for the function f (Table 6.47).

186

Coefficient NRG*ZRG NRG*HDDAVG NRG*ROCKDEPTH NCIG*ZCIG R2

Model Additive 0.221 (4.79) 0.149 (6.97) -1.429 (-6.98) 0.102 (2.32) 0.824

Table 6.47 Measurement and Regulation Station Cost Function Estimates (n=135)

The cost function is reformulated (Eq. 6.40) to compute correct elasticities (6.41-6.45):

(6.40)

(6.41)

(6.42)

(6.43)

(6.44)

(6.45) 187

The elasticity analysis (Table 6.48) shows that the average heating degree days variable has the strongest effect on costs: a 1% increase in this variable increases costs by almost 1.9%. Average rock depth is the distance between top soil and the rock layer. The bigger the depth, the lesser the effort needed for excavation. A 1% increase in average soil rock depth leads to 1.663% decrease in costs. A 1% increase in the number of residential customers and residential sales increases costs by 0.227% and 0.462%, respectively. Commercial-industrial sales have a small effect on costs (0.0002), since measurement and regulation stations serve primarily small commercial customers. The larger customers are served by industrial measurement and regulation stations, which are discussed in the following section.

Elasticity εNRG εSRG εSCIG εHDDAVG εROCKDEPTH

Sample Mean 0.227 0.462 0.0002 1.890 -1.663

Table 6.48 Measurement and Regulation Stations Cost Elasticities at the Sample Mean (n=135)

188

6.3.4 Cost Model for Industrial Measurement and Regulation Stations

Industrial measurement and regulation stations regulate gas pressure for industrial customers. These stations are located above ground, and are closely related to commercial-industrial characteristics. The selected model is:

= NCIG f(ZCIG, SOILSTEEL)

(6.46)

where NCIG = Number of commercial-industrial gas customers (#) ZRG = Commercial-industrial gas customer size (Commercial-industrial gas sales (mcf)/Number of commercial-industrial customers (#)) SOILSTEEL = Share of soil susceptible to steel corrosion (%)


Variable

($) NCIG (#) ZCIG (mcf/#) SOILSTEEL (%)

Minimum 544 2 23 0.39

Maximum 126,410 3,436 13,808 84.59

Mean 29,346 358 1,964 30.88

Std. Deviation 32,832 609 2,905 18.71

Table 6.49 Descriptive Statistics for the Industrial Measurement and Regulation Stations Cost Model (n=41)

189

Data is available for only two companies: Central Hudson and Orange and Rockland. Therefore the sample is limited to 41 districts. A linear additive form is used for the function f. The results are presented in Table 6.50. The effect of customer size is nonlinear, whereas steel corrosivity has a positive effect on costs.

Coefficient NCIG*ZCIG NCIG*ZCIG2 NCIG*SOILSTEEL R2

Model Additive 0.073 (4.01) -6.0E-06 (-1.85) 0.503 (2.03) 0.733

Table 6.50 Industrial Measurement and Regulation Stations Cost Function Estimates (n=41)

The cost model is reformulated (Eq. 6.47), in order to compute the correct elasticities (Eqs. 6.48-6.50).

(6.47)

(6.48)

190

(6.49)

(6.50)

The elasticity results (Table 6.51) shows that the number of commercial-industrial sales has the greatest effect on costs: a 1% increase in this variable increases costs by 0.73%. Commercial-industrial customers have a positive effect, since a 1% increase in customers increases costs by 0.26%. Steel corrosivity has a positive effect on costs: a 1% increase in steel corrosivity increases costs by 0.2%.

Elasticity εNCIG εSCIG εSOILSTEEL

Sample Mean 0.265 0.735 0.200

Table 6.51 Industrial Measurement and Regulation Stations Cost Elasticities at the Sample mean (n=41)

6.4 Marginal Costs and Elasticities in Total Capital Investment

The sum of the component-related capital investment cost functions is equal to the total capital investment cost function. Using this summary function, total marginal costs and elasticities can be estimated for the various outputs. 191

6.4.1 Total Capital Investment: Electricity

Total electricity capital costs are the sum of: overhead conductors, CO, underground conductors, CU, conduits, COND, services, SERVE, poles, PO, transformers, TR, street lighting, STL, and station equipment, STEQ investments. A sample of 125 districts is selected, that provides data for all the variables. The total electricity capital investment cost function is estimated at this sample mean. The characteristics of the selected sample are presented in Table 6.52.

The total cost, CETOT, is then:

CETOT = FCO(NRE, SRE, DENSA, SLP_P5PL, INCAGG, HOUSDETATT ,LFE) + FCU(SRE, SCIE, DENSBLTP, VAL, STO_P13PL, LFE) + FCOND(CCU, DENSA, INCAGG, STO_P13PL, AGE_P40PL) + NRE*fSERVE(ZRE, DENSBLTP,SOILSTEEL) + NCIE*gSERVE (ZCIE, DENSBLTP,SOILSTEEL) + FPO(

,

, DENSBLTP, WPEE)

NL*fSTL(ZL, STR, DENSBLTP)

+ FTR(CCO, CCU, SRE, SCIE, SOILCORR, SOILWORK, LFE) +

+ FSTEQ(SRE, SCIE, LFE)

192

(6.51)

Variable

Definition

Sample Mean 12,111

NRE

Number of residential electricity customers (#)

SRE

Residential electricity sales (kwh)

NCIE

Number of commercial-industrial electricity customers (#)

SCIE

Commercial-industrial electricity sales (kwh)

NL

Number of lighting customers (#)

SL

Lighting sales (kwh)

DENSA

Area density (population/total square mile area)

2,961

DENSBLTP VAL

Builtup density (population/square mile builtup area)

4,352

INCAGG

Aggregate household income ($)

76,986,906 1,361 168,204,326 40 3,061,610

447,169

Median house value ($)

294,290,446

HOUSDETATT Number of detached and single-unit attached houses (#)

8,884

AGE_P40P L STO_P13P L

Share of houses 40 years and older

0.37

Share of buildings with 13 and more stories

0.02

SOILCORR

Soil corrosivity (%)

64.97

SOILWORK SOILSTEEL

Soil workability (%)

48.23 40.21

SLP_P5P L

Steel corrosivity (%) Share of the land area with slopes 5% and more

LFE

Electricity load factor (average hourly kWh sales/hourly kWh peak load)

WPEE

0.18 0.56 16,988

Wage per electricity employee ($)

Table 6.52 Characteristics of the Electricity Sample (n=125)

The total electricity capital cost estimated at the sample mean is the sum of the estimated component costs (Table 6.53). Overhead and underground conductors make up the largest cost shares, 22% and 18%, respectively. Services make up the lowest share (4%). Conduits, services and poles have cost shares less than 10%. Transformers, street lighting, and station equipment have very close cost shares, around 13%.

193

Component Definition

Capital Investment Cost ($)

Cost Share

CO

Overhead conductors

4,204,787

0.22

CU

Underground conductors

3,542,419

0.18

COND SERVE

Conduits

1,618,085

0.08

693,916

0.04

PO

Poles

1,752,118

0.09

TR

Transformers

2,316,111

0.12

STL

Street lighting

2,486,761

0.13

STEQ CTOTE

Station equipment

2,755,671

0.14

Services

Total System

19,369,867

Table 6.53 Estimated Components and Total Capital Costs.

6.4.2 Marginal Capital Costs: Electricity

Marginal capital investment costs are computed with respect to the following outputs: (1) number of residential electricity customers, NRE, (2) number of commercialindustrial electricity customers, NCIE, (3) residential electricity sales, SRE, (4) commercial-industrial electricity sales, SCIE, (5) number of lighting customers, NL, and (6) lighting sales, SL, are computed with the sum of partial derivatives of component cost functions with respect to each output variable.

(6.52)

194

(6.53)

(6.54)

(6.55)

(6.56)

(6.57)

The marginal costs for output Xi in the Box-Cox model is:

(6.58) 195

In the additive model, λ=θ=1; in the log-log model, λ=θ=0. Therefore the marginal cost for output Xi in the additive model is equal to output

in the log-log model is equal to

and the marginal cost for

.

The marginal costs of each output with regard to each component are estimated at the sample mean (Table 6.54). Commercial-industrial electricity customers have the highest marginal cost, $295.5/customer. Residential electricity customers have the second highest marginal cost, $212.2/customer. Adding a commercial-industrial customer costs almost 40% more than adding a residential customer, and 5.9 times more than adding a lighting customer ($49.5). Marginal cost with regard to C-I sales are much smaller than the corresponding residential marginal cost, reflecting well-known differences in peak loads, hence in capacity costs. Lighting sales have the highest marginal cost among the three sectors: $0.83/kwh. The marginal costs of built-up density and area density are $277 and -$88. Increasing the density of built-up areas provides savings three times higher than those obtained by increasing the overall district density. This is reasonable, as most of the network investments are concentrated in the built-up areas, where most of the customers are located.

196

Marginal Definition Cost

MC CO MC CU

Overhead conductors Underground conductors

MC COND Conduits

Number of Residential residential electricity electricity sales customers NRE

SRE

115.33

0.02

__

0.02

__

-66.896

__

0.001

__

__

-64.538

25.32

0.01

__

0.02

0.0004 __

-150.47 __

-90.20971

__

Transformers Station equipment

Total

__

__

__

MC TR

MC TOTE

__

0.003

0.01

Street lighting

0.003

__

__

29.17

MC STL

__

__

0.03

Poles

MC SERVE Services

__

__

MC PO MC STEQ

Number of Commercial- Number of Lighting Builtup Area commercial- industrial lighting sales density density industrial electricity customers electricity sales customers NCIE SCIE NL SL DENSBLTP DENSA

95.734

-12.255

-33.038

__

__

__

__

__

__

-16.53

__

42.34

0.00

295.50

0.004 __

__

__

__

__

49.49

0.83

-26.34

__

212.16

0.13

295.50

0.01

49.49

0.83

-276.76

-87.77

Table 6.54 Estimated Component Marginal Capital Costs at the Electricity Sample Mean ($)

6.4.3 Economies of Scale: Electricity

The capital cost elasticities of outputs are next estimated for each component at the sample mean (Table 6.55). The total elasticities (last column in Table 6.55) are computed by weighting each component elasticity by the component cost share (Table 6.53), and summing up the weighted elasticities. The total elasticity results indicate very slight diseconomies of scale at the sample mean, εC=1.015; while economies are gained from densification, with εD=0.934.

197

Overhead Underground Conduits Services Poles Transformers Street Station Total conductors conductors lighting equipment Elasticity

Elasticity Number of residential electricity customers: Residential electricity sales: Number of commercialindustrial electricity customers: Commercialindustrial electricity sales: Number of lighting customers: Lighting sales: Builtup density: Area density:

εNRE

0.072

__

__

0.026 0.136

0.014

__

__

0.249

εSRE

0.099

0.097

0.045

0.032 0.026

0.080

__

0.093

0.471

εNCIE

__

__

__

__

__

__

__

0.021

εSCIE

__

0.024

0.011

__ 0.005

0.066

__

0.037

0.139

εNL

__

__

__

__

0.004

__

0.004

0.021

__

__

εSL εDENSBLTP εDENSA

0.131 __

-0.020

-0.009

-0.023

__

0.015

-0.003 -0.027 __

__

0.131

-0.002

-0.006

__

-0.068

-0.005

__

__

-0.013

Table 6.55 Elasticities, and Ray Economies of Scale and Density at the Electricity Sample Mean (n=125)

6.4.4 Total Capital Investment: Natural Gas

Total natural gas capital costs are calculated as the sum of system component costs: mains, MAIN, services, SERVG, measurement and regulation stations, MRS, and industrial measurement and regulation stations, INDMRS. The total cost, CGTOT, is then:

198

CGTOT = FMAIN(SRG, SCIG, HOUSDET, WATDEPTH, WPEG, HDDAVG) + NRG fSERVG(ZRG, DENSBLTP, HOUSDET, SOILSTEEL,WATDEPTH) SOILSTEEL, WATDEPTH)

+ NCIG gSERVG (ZCIG, DENSBLTP, HOUSDET,

+ NRG f(ZRG, HDDAVG, ROCKDEPTH) + NCIG gMRS(ZCIG, HDDAVG, ROCKDEPTH)

+ NCIG fINDMRS(ZCIG, SOILSTEEL)

(6.59)

A sample of 33 districts is selected, that provides data for all the variables. All calculations are done at the sample mean. The characteristics of the selected sample (n=33) are presented in Table 6.56.

Variable

Definition

NRG

Number of residential natual gas customers (#)

SRG

Residential natural gas sales (mcf)

NCIG

Number of commercial-industrial natural gas customers (#)

SCIG

Commercial-industrial natural gas sales (mcf)

DENSBLTP HOUSDET

Builtup density (population/square mile builtup area)

HDDAVG

Average heating degree days (°F)

WATDEP TH

Water table depth (inches)

385,461 428 379,630 10,656 3,272

Number of detached houses (#)

521 3.79 43.75

ROCKDEP TH Average rock depth (inches) SOILSTEEL Steel corrosivity (%) WPEG

Sample Mean 4,687

29.05 15,805

Wage per natural gas employee ($)

Table 6.56 Characteristics of the Natural Gas Sample (n=33)

The total natural gas capital cost is the sum of the component costs estimated at the sample mean (Table 6.57). The cost shares indicate that mains, the basic component of the natural gas network, make up 76% of the cost of the system. Service costs make up

199

around one fifth of the total cost. Industrial measurement and regulation stations represent very small cost shares.

Component Definition

Capital Investment Cost ($) Cost Share (% )

MAIN SERVG

Mains

4,944,613

0.76

Services

1,352,662

0.21

MRS

Measurement and regulation stations

194,458

0.03

INDMRS CTOTG

Industrial measuement and regulation stations

31,946

0.005

Total System

6,523,679

Table 6.57 Estimated Components and Total Capital Costs.

6.4.5 Marginal Capital Costs: Natural Gas

Marginal capital investment costs are computed with respect to the following outputs: (1) number of residential natural gas customers, NRG, (2) number of commercialindustrial natural gas customers, NCIG, (3) residential natural gas sales, SRG, and (4) commercial-industrial natural gas sales, SCIG, with:

(6.60)

(6.61)

200

(6.62)

(6.63)

The results are presented in Table 6.58. Residential natural gas customers have a marginal cost of $89.5, about 4.6 times the marginal cost of adding a commercialindustrial customer. The marginal cost of residential sales is also almost 4.4 times the marginal cost of commercial-industrial sales. Both differences reflect differences in peak load and resulting capacity costs. Density has a positive marginal cost, because of services.

Marginal Cost

Definition

MC MAIN

Mains

__

6.34

__

1.17

__

MC SERVG

Services Measurement and regulation stations Industrial measuement and regulation stations

74.41

2.01

__

0.60

30.46

15.05

0.22

__

0.10

__

__

__

19.34

0.06

__

Total

89.46

8.57

19.34

1.93

30.46

MC MRS MC INDMRS

MC TOTG

Number of Residential Number of commercialCommercialBuiltup residential natual natural gas industrial natural gas industrial natural density gas customers sales customers gas sales NRG SRG NCIG SCIG DENSBLTP

Table 6.58 Estimated Component Marginal Capital Costs for the Natural Gas Sample ($)

201

6.4.6 Economies of Scale: Natural Gas

The capital cost elasticities of outputs are estimated for each component at the sample mean (Table 6.59). The results show that at the sample mean, there are very slight economies of scale, εC=0.982, and almost a constant economies of density εD=0.993.

Elasticity Number of residential natural gas customers: Residential natural gas sales: Number of commercialindustrial natural gas customers: Commercialindustrial natural gas sales: Builtup density: Ray Economies of Scale Ray Economies of Density

Measurement and Industrial measuement Total regulation stations and regulation stations Elasticity

Mains

Services

εNRG

__

0.029

0.007

__

0.036

εSRG

0.354

0.109

0.002

__

0.464

εNCIG

__

__

__

0.00001

0.00001

εSCIG

0.395

0.070

0.002

0.004

0.471

__

0.011

__

__

0.011

0.749

0.218

0.011

0.004

0.982

0.749

0.228

0.011

0.004

0.993

εDENSBLTP

Table 6.59 Elasticity, and Ray Economies of Scale and Density in Natural Gas Sample Mean (n=33)

202

6.5 Multi-Utility Cost Analysis of Underground Network Investments

The economies of scope analysis in Chapter 5 (Section 5.2) is extended here by focusing on underground electricity and natural gas investments. There are 48 districts serving only electricity, and 186 serving both electricity and gas. The total underground investment cost is modeled as a function of total electricity and natural gas sales. A dummy variable is added to distinguish electricity only (DE) districts. Built-up area and low-rise housing reflect the effect of urban form on costs. Soil corrosivity is expected to increase both underground line and pipe investments. Heating degree days, an indicator of the need for more space heating due to cold weather, is also expected to increase costs, particularly in the case of natural gas. The selected model is:

CSi = F(STE, STG, DE, ABLTP, SOILCORR, STO_P1to3, HDDAVG) where STE = Total electricity sales (kWh) STG = Total natural gas sales (mcf) DE = Dummy variable for electricity-only districts ABLTP = Built-up area (square miles) SOILCORR = Soil corrosivity (%) STO_P1to3 = Share of buildings with 1 to 3 stories (%) HDDAVG= Average heating degree days (°F)

Descriptive statistics for above variables are presented in Table 6.60. 203

(6.64)

Variable CSCU ($) STE (kWh) STG (mcf) AREABLTPD (sq. miles) SOILCORR (%) STO_P1to3 (%) HDDAVG (°F)

Minimum 944 304,336 142 0.04 0.03 0.64 454

Maximum 159,386,294 3,113,865,373 13,035,016 133.85 0.98 0.99 731

Mean 7,117,253 142,642,666 667,972 5.98 0.78 0.96 524

Std. Deviation 17,481,316 372,108,758 1,437,502 13.09 0.27 0.05 56

Table 6.60 Descriptive statistics for the underground multi-utility investment cost model (n=234)

The Box-Cox form is superior to the log-log form (Table 6.61). All the variables are significant, with the expected signs. Increasing the service area has a positive effect on costs, whereas the share of low-story housing has a negative effect. Low-rise, lowdensity areas have usually larger roads, which make excavation operations easier, and thus decrease costs. Soil corrosivity increases costs, since corrosion requires special precautions and extra coating. HDDAVG requires more capacity investment, particularly natural gas plant capacity. Economies of scope are analyzed with scope scores5. Scope scores for different levels of electricity and natural gas outputs are calculated while holding site-specific and geographic variables constant at their sample mean values (Table 6.62). All scope scores are positive, hence economies scope are evident at every output combinations (Figure 6.1). Scope scores decrease with increasing electricity and natural gas outputs. The highest level of economies of scope (0.539) is obtained when both sectors’ outputs are minimal (83 Mwh and 0.14 mmcf), whereas the lowest level of economies of scope

5

See Section 5.2, Eqs. 5.19 and 5.20 for the definition of economies of scope and scope scores.

204

(0.186) is obtained when both sectors’ outputs are maximal (3,113,865 Mwh and 13,035 mmcf).

Coefficient Constant STE STG DE ABLTPD SOILCORR STO_P1to3 HDDAVG

Models Log-log -11.541 (-1.92) b 0.538 (5.60) 0.135 (2.19) 1.887 (1.09) 0.263 (2.82) 0.088 (2.13) -3.484 (-2.30) 2.633 (2.85)

λ

0.236 (0.000) c 0.169 (0.000)

θ R2 Log-likelihood H0: θ=λ=0d, H0: θ=λ=1, a Selected model b t-statistics in parentheses c p-value in parentheses d Log-likelihood test results

Box-Cox(λ, θ)a -105.106 (-2.81) 0.099 (6.83) 0.092 (2.49) 11.033 (3.06) 2.325 (3.00) 1.258 (2.18) -45.548 (-2.93) 8.074 (3.35)

0.694

0.763

-3683.10 Chi-square = 53.55 Chi-square = 839.75

-3656.33 (0.000) (0.000)

Table 6.61 Multi-utility underground cost function estimates (n=234)

205

ELECTRICITY (1000 kwh)

NATURAL GAS (1000 mcf)

304 346,323 692,342 1,038,360 1,384,379 1,730,398 2,076,417 2,422,436 2,768,455 3,113,865 0.14 0.539

0.297

0.270

0.255

0.244

0.236

0.230

0.224

0.220

0.216

1,448 0.439

0.263

0.242

0.230

0.221

0.215

0.209

0.205

0.201

0.197

2,896 0.423

0.258

0.237

0.226

0.217

0.211

0.206

0.201

0.198

0.194

4,345 0.413

0.254

0.234

0.223

0.215

0.208

0.203

0.199

0.195

0.192

5,793 0.406

0.251

0.232

0.221

0.213

0.206

0.201

0.197

0.194

0.191

7,241 0.400

0.249

0.230

0.219

0.211

0.205

0.200

0.196

0.192

0.189

8,690 0.395

0.247

0.228

0.217

0.210

0.204

0.199

0.195

0.191

0.188

10,138 0.391

0.245

0.227

0.216

0.208

0.203

0.198

0.194

0.190

0.187

11,586 0.387

0.244

0.226

0.215

0.207

0.202

0.197

0.193

0.189

0.186

13,035 0.384

0.243

0.225

0.214

0.206

0.201

0.196

0.192

0.189

0.186

Table 6.62 Scope values for different levels of electricity and natural gas outputs

Figure 6.1 Scope scores for different levels of electricity and natural gas outputs.

Values of S are calculated for the same output levels, as in Table 6.62, while using the minimum and maximum values of site-specific variables: built-up area, soil corrosivity, share of low-rise houses, and average heating degree days. The

scores for

4 combinations of outputs and at the minimum/maximum values of the site-specific 206

variables are presented in Table 6.63. The complete sets of score values are presented in Appendix C, graphical representations of the scope surfaces are presented on Figure 6.2.

Residential Customers and Site-specific variables AB LTP SOILCORR STO1to3 HDDAVG

Output levels (electricity ; natural gas) (304 ; 0.14) (304 ; 13,035)

(3,113,865 ; 0.14)

(3,113,865 ; 13,035)

minimum

0.411

0.271

0.141

0.120

maximum

0.667

0.515

0.319

0.280

minimum

0.514

0.361

0.199

0.171

maximum

0.542

0.387

0.218

0.187

minimum

0.673

0.523

0.325

0.286

maximum

0.524

0.369

0.205

0.176

minimum

0.477

0.327

0.176

0.151

maximum

0.641

0.488

0.296

0.258

Table 6.63 Scope values for different levels of site-specific variables and different output combinations.

The results for soil corrosivity are very close, and similar to those presented in Table 6.62 and on Figure 6.1 (where the mean soil corrosivity variable is used). Economies of scope are highest when the share of low-rise houses is minimal, which suggests that such neighborhoods may provide more opportunities for efficient design of joint underground systems, hence for lower costs. Higher economies of scope are achieved for the maximum average heating degree days, suggesting that higher capacity needs facilitate the joint design of both underground networks, thus reducing costs. The overall lowest score (0.120) is obtained for the maximal outputs and smallest built-up area. The S score is close to its maximum value (0.667) when the outputs are minimal 207

and the built-up area maximal, which suggests that small markets over large territories provide more opportunity for savings from joint underground investments.

(a) Built-up area

(b) Soil corrosivity Figure 6.2 Scope scores for different values of electricity and natural gas sales, and for different levels of site-specific, geographic, and company-specific variables: (a) Built-up area, (b) Soil corrosivity, (c) Average heating degree days, (d) Share of buildings with 1 to 3 stories 208


(c) Average heating degree days

(d) Share of buildings with 1 to 3 stories

209

6.6 Summary

The disaggregate analysis of electricity and natural gas distribution capital costs shows that system components may display characteristics different from those of the whole system. Some of the variables, such as residential and commercial-industrial sales and density, are common to component cost models. However, each component has a particular cost structure and economic characteristics. Geographic variables such as slope, soil corrosivity, steel corrosivity, and workability; site-specific characteristics such as the shares of low-rise and high-rise buildings, the age of housing, and the wealth of the area; and company-specific characteristics, such as load factor and wage per employee, affect different components in different ways. The Box-Cox model provides a better fit for network components: overhead and underground conductors, and conduits in electricity system; and mains in natural gas system. The linear additive model is used for end-user related components: services and street lighting in electricity systems; and services, and measurement and regulation stations in natural gas systems. In the electricity system, overhead and underground conductors are the backbone of the network. Overhead and underground conductors have been further disaggregated as primary and secondary lines, but the availability of primary and secondary lines data is limited to two companies (Central Hudson and LILCO), and therefore the results may display characteristics different from those for the total lines. For instance, overhead conductors enjoy economies of scale and density at the sample mean (n=241), whereas primary overhead conductors experience slight economies of scale and economies of 210

density (n=137), and secondary overhead conductors appear to experience diseconomies of scale and density (n=137). The wealth of the area has different effects on different components. Income has a negative effect on overhead conductor costs, whereas median house value has a positive effect on underground conductor costs, which may be the result of an increasing willingness by customers to pay for the aesthetics and safety of underground systems. Density has also different effects on different components. Density has a negative effect on the costs of overhead and underground conductors, which is likely due to less and shorter lines, whereas it has a positive effect on conduit costs, which may be related to congestion expenses. Electricity services and street lighting are estimated with linear additive models. The number of residential customers and residential customer size, commercial-industrial sales, built-up density and steel corrosivity are the variables affecting service costs, while lighting sales, lighting customer size, total street length and built-up density are the variables affecting street lighting costs. Conduits, poles and transformers are directly related to other system components. Conduit costs are related to underground conductors, transformers are related to overhead and underground conductors, and poles are related to overhead conductors. The Box-Cox form provides the best fit for the conduit, pole, and transformer cost models. In the natural gas system, mains are the backbone of the system. The cost function for mains is estimated using the Box-Cox form. Residential and commercial-industrial sales, the number of detached houses, water table depth, wage per employee and average heating degree days are the variables affecting main costs. Mains enjoy economies of 211

density. When the number of detached houses is taken as an indicator of the service area size, mains experience slight economies of scale. Natural gas servives, measurement and regulation stations, and industrial measurement and regulation stations are end-user related components, and are estimated using a linear additive form. The number of residential customers, residential customer size, commercial-industrial sales affect natural gas service costs, which are also nonlinearly related to density. The number of detached houses, steel corrosivity and water table depth are the other variables affecting service costs. Measurement and regulation stations (MRS) serve residential and small commercial customers, and therefore their costs are taken as a function of the number of residential customers, residential customer size, and commercial-industrial customer size. Average heating degree days and average rock depth are the other variables affecting station costs. Industrial MRS information is limited to two companies: Central Hudson, and Orange and Rockland. This component cost is estimated as a function of the number of commercial-industrial customers, commercial-industrial customer size and steel corrosivity. The next analysis has considered the marginal costs and elasticities for the total capital investment. Component costs are summed up to obtain the total capital investment costs. Two samples of tax districts are used for this analysis: 125 tax districts for electricity, and 33 tax districts for natural gas. The electricity marginal cost analysis shows that an additional commercialindustrial customer is 1.4 times more costly than an additional residential customer, and 5.9 times more costly than an additional lighting customer. The marginal costs of sales 212

are highest for lighting customers and smallest for commercial-industrial customers. Increasing density has negative marginal costs varying between $87 and $276, depending on the type of geographic unit: tax district area vs. built-up area. Slight economies of scale, and stronger economies of density take place at the sample mean. The natural gas marginal cost analysis shows that adding a residential gas customer incurs a cost of $89, versus $19 for a commercial-industrial customer. The marginal cost of residential sales ($8.64/mcf) is about 4.6 times the marginal cost of commercial-industrial sales. The built-up density has a positive marginal cost of $30. Very slight economies of scale and almost constant returns to density are observed at the natural gas sample. Finally, a multi-utility cost analysis has been conducted on underground electricity and natural gas investments. 186 tax districts serve both electricity and natural gas, and 48 districts serve electricity only. Underground investment costs are estimated as a function of total electricity and natural gas sales, built-up area, soil corrosivity, share of low-rise houses, and average heating degree days. The Box-Cox form turned out superior to the linear and log-log forms. Scope scores are computed to find the levels of economies of scope for different output combinations. Scope scores vary between 0.186 and 0.539, with scores decreasing with increasing outputs. The positive score values for all combinations provide evidence for the existence of economies of scope. Scope scores are also computed while considering different values of the site-specific variables. The results vary between 0.120 and 0.673, again providing evidence for economies of scope. The overall highest score is observed at minimal outputs and minimal share of low-rise

213

houses, whereas the overall lowest score is observed at the maximal outputs and smallest built-up area. In summary, output variables, density and various site-specific and companyspecific variables do affect component costs of electricity and natural gas systems. The main system components: overhead and underground conductors in electricity, and mains in natural gas, display economies of scale and economies of density. However, diseconomies of scale and density are observed in some other components such as services and regulation stations. Economies of scope take place over all ranges of independent variables, supporting the results in Section 5.2.

214

CHAPTER 7 CONCLUSIONS

Past empirical research on electricity and natural gas distribution cost modeling has focused primarily on output variables such as numbers of customers and sales, density, company-specific characteristics, and a very limited number of site-specific characteristics in electricity and natural gas investment cost estimations. This research has aimed to expand such by considering various site-specific urban and geographic variables and socio-economic characteristics. The existence economies of scale, density economies, and scope have been investigated. This research has only considered capital investment costs. Prior to the cost analyses, a market analysis has conducted with the goal of expanding the available company data. However, the larger dataset by the non-price models failed to yield reasonable estimates for the investment cost analyses, which were then focused only on the communities with actual company data. Capital investment costs are analyzed with both aggregate and disaggregate electricity and natural gas distribution data. The number of residential customers, residential and commercial-industrial sales, area density, soil corrosivity, number of street intersections, and electricity load factor turn out to be significant determinants of 215

electricity distribution costs. The Box-Cox form was superior to the log-log and linear forms in aggregate cost analyses. Economies of scale and density were observed for different average customer sizes, but economies obtained from densification were stronger than those obtained with output expansion. The elasticity curves for the minimum customer size has horizontal asymptotes, at εCE=0.961 and εDE=0.859, which implies that there are never diseconomies of scale and diseconomies of density. Scale and density analyses were also conducted for different levels of site-specific, geographic, and company-specific variables. The results provided evidence for economies of scale and density for different levels of these variables. Natural gas distribution costs were estimated using the number of residential customers, residential and commercial-industrial gas sales, built-up area, area share of workable soil, number of street intersections, and average wage per employee. The BoxCox form was superior to the log-log and linear forms in the aggregate cost analyses, in a way similar to electricity distribution costs. Economies of scale and density were observed for different average customer sizes. Scale and density analyses were expanded to different levels of site-specific, geographic, and company-specific variables, and economies of scale and density were observed for all the values of these variables. However, point elasticities calculated for each observation (district), indicated that three districts experience very slight diseconomies of scale. Using the actual data of these tax districts, the trade-offs between soil workability and the number of street intersections are analyzed in Appendix B. Economies of scope were analyzed for aggregate multi-utility distribution costs. The cost function included total electricity and natural gas sales, number of street intersections, and share of 40+ year-old houses. The Box-Cox form was 216

superior to the log-log and linear forms. The results indicated the presence of economies of scope for various output combinations, and scope scores decreased with increasing outputs. Economies of scope analyses were also conducted for different levels of street intersections and share of old houses, and economies of scope were observed in all levels. The next analyses considered disaggregate capital investment costs. Electricity component cost models were estimated for: overhead conductor (primary and secondary), underground conductor (primary and secondary), conduit, service, pole, transformer and station equipment. In all these models, the Box-Cox form turned out to be superior to the log-log and linear models, except for services and street lighting. Services and street lighting are end-user related component. In this case, investment costs are taken as linear functions of the number of customers and their characteristics. Overhead conductors have been estimated as a function of the number of residential customers and sales, area density, share of land area with slopes of 5% and more, aggregate income, number of detached and attached houses, and electricity load factor. The results indicated that increasing outputs have positive effects on costs. Steep slopes, more single houses also have positive effects on costs, whereas increasing income and load factor have negative effects. Economies of scale and density were observed for this component. Overhead conductors were further disaggregated into primary and secondary conductors, with data available for only 2 companies. Residential and commercial-industrial sales and built-up density were influential on primary overhead conductor costs. The number of residential customers and built-up density were also influential on overhead conductor costs, in addition to steep slopes and the share of detached houses. Economies of scale and density were observed for primary overhead 217

conductors, whereas diseconomies of scale and density were observed for secondary overhead conductors. Underground conductors costs were estimated as a function of residential and commercial-industrial sales, built-up density, median house value, share of buildings with 13 and more stories, and electricity load factor. Wealth variable had a positive effect on underground conductor costs, unlike overhead conductor costs, which indicates that wealthier neighborhoods have a higher willingness to pay for underground lines due to aesthetic and safety preferences. Underground conductors were further disaggregated into primary and secondary lines, but the data were available for only two companies. Primary and secondary line costs have variables very similar to those of the aggregate underground conductor costs, with the inclusion of the share of houses 40+ years-old. Primary underground conductors experience diseconomies of scale and economies of density, while secondary underground conductors enjoy economies of scale and density. Conduits have been modeled under two alternative approaches. The first considers aggregate underground conductors investments, in addition to area density, aggregate income, share of buildings with 13 and more stories, and share of houses 40+ years-old, as explanatory variables. The second considers primary and secondary underground conductors, using a smaller sample size. The area density and the water table depth turned out as additional explanatory variables. The density variables have positive effects on costs, unlike the other models, which might result from increasing congestion prices. Income had a positive effect on conduit costs, in a way similar to underground conductor costs.

218

Services have been estimated as a function of the numbers of residential and commercial-industrial customers, residential and commercial-industrial customer sizes, built-up density and steel corrosivity. Poles were closely related to overhead conductors, with built-up density and average wage per employee also influential on pole investment costs. Transformers were closely related to conductors. Transformer investment costs were estimated as a function of overhead and underground conductors, residential and commercial-industrial sales, soil corrosivity and soil workability, and electricity load factor. Street lighting costs turned out to be affected by the number of lighting customers, and lighting customer average size, in addition to total street length and builtup density. The last component for electricity, station equipment, was estimated as a function of residential and commercial-industrial sales, and load factor. Natural gas system components include: mains, services, measurement and regulation stations (MRS), and industrial MRS. Mains are the backbone of the system. Main investment costs were estimated using the Box-Cox form, since it is superior to the log-log and linear forms. The other three components are end-user related components, and were estimated using linear-additive forms, in a way similar to services and street lighting in electricity. Main investment costs were estimated as a function of residential and commercial-industrial sales, number of detached houses, water table depth, wage per employee, and average heating degree days. Mains are underground investments, and therefore it is plausible for the average wage to increase costs, and for the water table depth to decrease costs. Higher average heating degree days require more main

219

investments, since natural gas is used generally for house heating. Mains enjoyed economies of scale and density. Service investment costs were estimated as a function of the number of residential customers and residential customer size, commercial-industrial sales, built-up density, number of detached houses, steel corrosivity, and water table depth. Built-up density has a non-linear relationship with investments, due to the use of density squared. Steel corrosivity and water table depth are expected to be related to underground service lines. The number of residential customers, the average residential customer size, and commercial-industrial sales, in addition to average heating degree days and average rock depth had effects on MRS investment costs. Industrial MRS was only related industrial customers, and therefore their costs were estimated using the number of commercialindustrial customers, commercial-industrial average customer size and steel corrosivity. The individual component investment cost estimations were followed by marginal costs and elasticities analyses for the total capital investment, taken as the sum of the component investment costs. Two samples were selected for electricity and natural gas, where all variables were available. Density had negative marginal costs for electricity, indicating savings. In natural gas, on the other hand, density had positive marginal costs. Electricity sample experienced slight diseconomies of scale and slight economies of density at the mean, while natural gas sample enjoyed slight economies of scale and density at the mean. The last analysis considered multi-utility investment costs for underground system components: underground conductors and conduits for electricity, and mains for natural gas. The cost function included total electricity and natural gas sales, built-up 220

area, soil corrosivity, share of buildings with 1 to 3 stories, and average heating degree days. The results provided evidence for economies of scope in underground network at different output combinations, and scope scores decreased with increasing outputs. The scope analysis was expanded using different values for site-specific variables, and economies of scope were always observed. In summary, various socio-economic and site-specific urban and geographic factors appear to influence electricity and natural gas distribution investment costs, in addition to output and company-specific variables. Electricity and natural gas distribution investments generally enjoy economies of scale and density. Economies of scope were evident for these sectors. Therefore, investing on electricity and natural gas appeared to be economically more feasible than investing on each sector separately. The results may provide insights into policy making for urban development. The selection of future development sites can be analyzed in terms of environmental and geographic characteristics, and the density of urban development patterns must be considered in detail, since all these factors are likely to impact energy infrastructure investment costs. There are several ways to expand this research. First, economies of scope analyses can be conducted with further disaggregation of the output variables. Instead of total customers and sales, the number of customers and sales for each sector could be considered to investigate the effects of different sectors. Second, a company efficiency analysis could be conducted using non-parametric (DEA) and parametric (SFA and COLS) techniques. Finally, marginal cost analyses for total capital investments could be expanded by considering different local markets for electricity and natural gas, with implications for spatialized pricing of these products. 221

BIBLIOGRAPHY

Bagdadioglu, N., C. M. W. Price, and T. G. Weyman-Jones. 1996. Efficiency and Ownership in Electricity Distribution: A non-parametric model of the Turkish experience. Energy Economics, vol. 18, pp. 1-23. Baumol, W. J., J.C. Panzar, and R. D. Willig. 1982. Contestable Markets and the Theory of Industry Structure, Harcourt Brace Jovanovich, New York. Bernard, J., D. Bolduc, and A. Hardy. 2002. The Costs of Natural Gas Distribution Pipelines: the case of SCGM, Quebec. Energy Economics, vol. 24, pp. 425-438. Bhattacharyya, A., T. R. Harris, R. Narayanan, and K. Raffiee. 1995. Allocative Efficiency of Rural Nevada Water Systems: A Hedonic Shadow Cost Function Approach. Journal of Regional Science, vol. 35, no. 3, pp. 485-501. Box, G. E. P. and D. R. Cox. 1964. An Analysis of Transformations. Journal of the Royal Statistical Society, Series B, vol.26, no. 2, pp. 211-252. Burns, P. and T. G. Weyman-Jones. 1996. Cost Functions and Cost Efficiency in Electricity Distribution: A Stochastic Frontier Approach. Bulletin of Economic Research, vol. 48, no. 1, pp. 41-64. Casarin, A. A. 2007. Efficient Industry Configurations in Downstream Gas Markets: An empirical assessment. Energy Economics, vol. 29, pp. 312-328. Chambers, W. L. 1959. Engineering Soil Classification for Residential Developments. Federal Housing Administration, Washington D.C.

222

Chapell, H., W., Jr., R. P. Wilder, and J. W. Mayo. 1986. Multiproduct Monopoly, Regulation, and Firm Costs: Comment. Southern Economic Journal, vol. 52, no. 4, pp. 1168-1174. Chen, T. 2002. An Assessment of Technical Efficiency and Cross-efficiency in Taiwan’s Electricity Distribuion Secor. European Journal of Operational Research, vol. 137, pp. 421-433. Christiensen, L. R. and W. H. Greene, 1976. Economies of Scale in U.S. Electric Power Generation. The Journal of Political Economy, vol. 84, no. 4(1), pp. 655-676. Claggett, E. T. Jr. 1994. A Cost Function Study of the Providers of TVA Power. Managerial and Decisions Economics, vol. 15, no. 1, pp. 63-72. Claggett, E. T. Jr., D. R.Hollas, and S. R. Stansell. 1995. The Effects of Ownership Form on Profit Maximization and Cost Minimization Behavior within Municipal and Cooperative Electrical Distribution Utilities. The Quarterly Review of Economics and Finance, vol. 35, pp. 533-550. Cubukcu, K. M. 2003. Geography and the Cost of Network Infrastructure: The Case of Local Telephone Systems, Unpublished Dissertation, The Ohio State University. Cubukcu, K. M., and Guldmann, J. 2008. Geography and the Costs of Local Telephone Networks. Annals of Regional Science, vol. 42, pp. 821-842. Durrer, E. J., and G. E. Slater. 1977. Optimization of Petroleum and Natural Gas Production – A Survey. Management Science, vol. 24, no. 1, pp. 35-43. EIA, Electricity Explained, last retrieved in February 2011 http://www.eia.gov/energyexplained/index.cfm?page=electricity_in_the_united_st ates.

223

EIA. 2008. Distribution of Natural Gas: The Final Step in the Transmission Process, Energy Information Administration, Office of Oil and Gas, June 2008, last retrieved in February 2011 http://www.eia.gov/pub/oil_gas/natural_gas/feature_articles/2008/ldc2008/ldc200 8.pdf. Erbetta, F., and L. Rappuoli. 2008. Optimal Scale in the Italian Gas Distribution Industry Using Data Envelopment Analysis. Omega International Journal Management Science, vol. 36, pp. 325-336. Erturk, M., and S. Turut-Asik. 2011. Efficiency Analysis of Turkish Natural Gas Distribution Companies by Using Data Envelopment Analysis Method. Energy Policy, vol. 39, pp. 1426-1438. Fabbri, P., and Fraquelli, G. 2000. Costs and Structure of Technology in the Italian Water Industry. Empirica, vol. 27, pp. 65-82. Fabbri, P., G. Fraquelli, and R. Giandrone. 2000. Costs, Technology and Ownership of Gas Distribution in Italy. Managerial and Decision Economics, vol. 21, pp. 71-81. Farsi, M., and M. Filippini. 2004. Regulation and Measuring Cost-Efficiency with Panel Data Models: Application to Electricity Distribution Utilities. Review of Industrial Organization, vol. 25, pp. 1-19. Farsi, M., Fetz, A., and Filippini, M. 2008. Economies of Scale and Scope in Multi Utilities. The Energy Journal, vol.29, no. 4, pp. 123-143. Farsi, M., M. Filippini, and M. Kuenzle. 2007. Cost Efficiency in the Swiss Gas Distribution Sector. Energy Economics, vol. 29, pp. 64-78. Farsi, M., M. Filippini, and W. Greene. 2006. Application of Panel Data Models in Benchmarking Analysis of the Electricity Distribution Sector. Annals of Public and Cooperative Economics, vol. 77, no. 3, pp. 271-290. Filippini, M. 1996. Economies of Scale and Utilization in the Swiss Electric Power Distribution Industry. Applied Economics, vol. 28, pp. 543-550. 224

___________. 1998. Are Municipal Electricity Distribution Utilities Natural Monopolies? Annals of Public and Cooperative Economics, vol. 69, no. 2, pp. 157-174. Filippini, M. and J. Wild. 2001. Regional Differences in Electricity Distribution Costs and Their Consequences for Yardstick Regulation of Access Prices. Energy Economics, vol.23, pp. 477-488. Filippini, M., J. Wild, and M. Kuenzle. 2001. Scale and Cost Efficiency in the Swiss Electricity Distribution Industry: Evidence from a Frontier Cost Approach. Centre for Energy Policy and Economics Working Paper Nr. 8. Centre for Energy Policy and Economics Swiss Federal Institutes of Technology, Zurich, Switzerland. Filippini, M., N. Hrovatin, and J. Zoric. 2004. Efficiency and Regulation of the Slovenian Electricity Distribution Companies. Energy Policy, vol. 32, pp. 335-344. Folloni, G. and O. Caldera. 2001. Size, Density and Costs of Network Services – The Case of the Distribution of Electricity in Italy. European Regional Development Issues in the New Millennium and Their Impact in Economic Policy, 41st Congress of the European Regional Science Association, Paper number 196, pp. 1-17. Fraquelli, G., Piacenza, M.,and Vannoni. 2004. Scope and Scale Economies in Multiutilities: Evidence from Gas, Water and Electricity Combinations. Applied Economics, vol. 36, pp. 2045-2057. Fraquelli,G., M. Piacenza, and D. Vannoni. 2005. Cost Savings from Generation and Distribution with an Application to Italian Electric Utilities. Journal of Regulatory Economics, vol. 28, no.3, pp. 289-308. Giannikis, D., T. Jamasb, and M. Pollitt. 2005. Benchmarking and Incentive Regulation of Quality of Service: an application to the UK electricity distribution network. Energy Policy, vol. 33, pp. 2256-2271. Gilsdorf, Keith. 1995. Subadditivity of Vertically Integrated Electric Utilities. Southern Economic Journal, vol. 62, no. 1, pp.126-138. 225

Goto, M., and M. Tsutsui. 1998. Comparison of Productive and Cost Efficiencies Among Japanese and US Electric Utilities. Omega International Journal Management Science, vol. 26, no. 2, pp. 177-194. Gribbin, T. K. 1953. Production Costs in the Gas Industry. Oxford Economic Papers, vol. 5, pp. 190-208. Guldmann, J. 1983. Modeling the Structure of Gas Distribution Costs in Urban Areas. Regional Science and Urban Economics, vol. 13, pp. 299-316. ___________. 1984. A Further Note on the Structure of Gas Distribution Costs in Urban Areas. Regional Science and Urban Economics, vol. 14, pp. 583-588. ___________. 1984a. An Econometric Model of Electricity Distribution Systems in Urban Areas. Environment and Planning A, vol. 16, pp. 793-806. ___________. 1985. A Disaggregate Econometric Analysis of Electricity Distribution Capital Costs. Energy, vol. 5, pp. 601-612. ___________. 1985a. Economies of Scale and Natural Monopoly in Urban Utilities: The Case of Natural Gas Distribution. Geographical Analysis, vol. 17, no. 4, pp. 302317. ___________. 1988. Land Use, Market Mix, and the Allocation of the Investment Costs of Electricity Distribution Networks. Socio-Economic Planning, vol. 22, no. 5, pp. 201-212. ___________. 1989. Capacity Cost Allocation in the Provision of Urban Public Services: The Case of Gas Distribution. Growth and Change, vol. 20,no. 2, pp. 1-18. Hattori, T., T. Jamasb, and M. Pollitt. 2005. Electricity Distribution in the UK and Japan: A Comparative Efficiency Analysis 1985-1998. The Energy Journal, vol. 26, no. 2, pp. 23-45.

226

Henderson, S. J. 1985. Cost Estimation for Vertically Integrated Firms: The Case of Electricity. In Michael A. C. (Ed.), Analyzing the Impact of Regulatory Change in Public Utilities, Lexington Books, Massachusetts, Toronto, pp. 75-94. Hjalmarsson, L., and A. Veiderpass. 1992. Productivity in Swedish Electricity Retail Distribution. Scandinavian Journal of Economics, vol. 94 Supplement, pp. 193205. Hollas, D., R. 1999. Gas Utility Prices in a Restructured Industry. Journal of Regulatory Economics, vol. 16, pp. 167-185. Hollas, D. R., and S. R. Stansell. 1994. The Economic Efficiency of Public vs. Private Gas Distribution Utilities. Annals of Public and Cooperative Economics, vol. 65, no. 2, pp. 281-300. Hollas, D. R., K. R. Macleod, and S. R. Stansell. 2002. A Data Envelopment Analysis of Gas Utilities’ Efficiency. Journal of Economics and Finance, vol. 26, no. 2, pp. 123-137. Huettner, D. A., and J. H. Landon. 1978. Electric Utilities: Scale Economies and Diseconomies. Southern Economic Journal, vol. 44, no. 4, pp. 883-912. Ibrahim, E. S. 1999. Corrosion Control in Electric Power Systems. Electric Power Systems Research, vol. 52, pp. 9-17. Ingco, S. P. 1996. Structural Changes in Power Sector in Asia: Improving Profitability. Energy Policy, vol.24, no.8, pp. 723-733. Jamasb, T., L. Orea, M. Pollitt. 2012. Estimating the Marginal Cost of Quality Improvements: The Case of the UK Electricity Distribution Companies. Energy Economics, vol. 34, pp. 1498-1506. Jamasb, T., and M. Pollitt. 2003. International Benchmarking and Regulation: an application to European electricity distribution utilities. Energy Policy, vol. 31, pp. 1609-1622. 227

Joskow, Paul, L., and Schmalensee, Richard. 1983. Markets for Power: An Analysis of Electric Utility Deregulation, The MIT Press, Cambridge, Massachusetts. Kaserman, D. L. and J. W. Mayo. 1991. The Measurement of Vertical Economies and the Efficient Structure of the Electric Utility Industry. The Journal of Industrial Ecnomics, vol. 39, no. 5, pp. 483-502. Kim, H. Y. 1987. Economies of Scale in Multi-Product Firms: An Empirical Analysis. Economica, vol. 54, no.214, pp. 185-206. Kim, T., and J. Lee. 1995. Cost Analysis of Gas Distribution Industry with Spatial Variables. Journal of Energy and Development, vol. 20, no. 2, pp. 247-267. Kinnunen, K. 2005. Pricing of Electricity Distribution: An Empirical Efficiency Study in Finland, Norway and Sweden. Utilities Policy, vol. 13, pp. 15-25. Klein, C. C. 1993. A Comparison of Cost-based Pricing Rules for Natural Gas Distribution Utilities. Energy Economics, vol. 15, no. 3, pp. 176-182. Kopsakangas-Savolainen, M., and R. Svento. 2008. Estimation of Cost-Effectiveness of the Finnish Electricity Distribution Utilities. Energy Economics, vol. 30, pp. 212229. Korhonen, P. J., and M. J. Syrjanen. 2003. Evaluation of Cost Efficiency in Finnish Electricity Distribution. Annals of Operations Research, vol. 121, pp. 105-122. Kwoka, J. E. Jr. 1996. Power Structure: Ownership, Integration, and Competition in the U.S. Electricity Industry, Kluwer Academic Publishers, Boston, Dordrecht, London. ___________. 2002. Vertical Economies in Electric Power: Evidence on Integration and Its Alternatives. International Journal of Industrial Organization, vol. 20, pp. 653671.

228

Longley, P. A., M. F. Goodchild, D. J. Maguire, and D. W. Rhind. 2005. Geographical Information Systems and Science, Wiley, Chichester. Maloney, M. T. 2001. Economies and Diseconomies: Estimating Electricity Cost Functions. Review of Industrial Organization, vol. 19, pp. 165-180. Mayo, J. W. 1984. Multiproduct Monopoly, Regulation, and Firm Costs. Southern Economic Journal, vol. 52, pp. 208-218. Mazumdar, M., and A. Kapoor. 1995. Stochastic Models for Power Generation System Production Costs. Electric Power Systems Research, vol. 35, pp. 93-100. Meyer, R. A. 1975. Publicly Owned versus Privately Owned Utilities: A Policy Choice. The Review of Economics and Statistics, vol. 57, no. 4, pp. 391-399. Miliotis, P. A.1992. Data Envelopment Anaysis Applied to Electricity Distribution Districts. Journal of the Operational Research Society, vol. 43, no. 5, pp. 549-555. Mork, K. A. 1992. Comment on L. Hjalmarsson and A. Veiderpass, “Productivity in Swedish Electricity Retail Distribution”. Scandinavian Journal of Economics, vol. 94 Supplement, pp. 207-209. Nelson, R. A., and Primeaux, Jr. W. J. 1988. The Effects of Competition on Transmission and Distribution Costs in the Municipal Electric Industry. Land Economics, vol. 64, no. 4, pp. 338-346. Nemoto, J., Y. Nakanishi, and S. Madono. 1993. Scale Economies and OverCapitalization in Japanese Electric Utilities. International Economic Review, vol. 34, no. 2, pp. 431-440. NOAA, Monthly Station Climate Summaries, last retrieved in August 2013, http://www.ncdc.noaa.gov/oa/documentlibrary/pdf/eis/clim20eis.pdf. Neuberg, L. G. 1977. Two Issues in the Municipal Ownership of Electric Power Distribution Systems. The Bell Journal of Economics, vol.8, no. 1, pp. 303-323. 229

Nunez, K. 2007. Electricity Utility Deregulation: Stranded Costs vs. Stranded Benefits. Journal of Accounting and Public Policy, vol. 26, pp. 193-211. Pacudan, R., E. De Guzman. 2002. Impact of Energy Efficiency Policy to Productive Efficiency of Electricity Distribution Industry in Philippines. Energy Economics, vol. 24, pp. 41-54. Palmer, J. D. 1989. Environmental Characteristics Controlling The Soil Corrosion of Ferrous Piping. In Chalker V. and J. D. Palmer (Eds.), Effects of Soil Characteristics on Corrosion, American Society for Testing and Materials, Ann Arbor, MI. Piacenza, M., and D. Vannoni. 2004. Choosing Among Alternative Cost Function Specifications: An Application to Italian Multi-utilities. Economics Letters, vol. 82, pp. 415-422. Porat, Y., I. Rotlevi, and R. Turvey. 1997. Long-run Marginal Electricity Generation Costs in Israel. Energy Policy, vol. 25, no. 4, pp. 401-412. Rao, P. and Miller R. L. 1971. Applied Econometrics, Wadsworth Publishing Company, Belmont, California. Rebak, R. B., Z. Xia, R. Safruddin, and Z. Szlarska-Smialowska. 1996. Effect of Solution Composition and Electrochemical Potential on Stress Corrosion Cracking of X-52 Pipeline Steel. Corrosion Science, vol. 52, no.5, pp. 396-405. Primeaux, W. J. Jr. 1975. A Re-examination of the Monopoly Market Structure for Electric Utilities. In Phillips, A. (Ed.) Studies in the Regulation of Economic Activity: Promoting Competition in Regulated Markets, The Brookings Institution, Washington D.C., pp. 175-200. Roberts, M. J. 1986. Economies of Density and Size in the Production and Delivery of Electric Power. Land Economics, vol. 62, no. 4, pp. 378-387. Rossi, M. A. 2001. Technical Change and Efficiency Measures: The Post-privatisation in the Gas Distribution Sector in Argentina. Energy Economics, vol.23, pp. 295-304. 230

Salvanes, K. G., and S. Tjotta. 1994. Productivity Differences in Multiple Output Industries: An Empirical Application to Electricity Distribution. The Journal of Productivity Analysis, vol. 5, pp. 23-43. ___________. 1998. A Test for Natural Monopoly with Application to Norwegian Electricity Distribution. Review of Industrial Organization, vol. 13, pp. 669-685. Schmalensee, R., P. L. Joskow. 1986. Estimated Parameters as Independent Variables: An Application to the Costs of Electric Generating Units. Journal of Econometrics, vol. 31, pp. 275-305. Sing, M. 1987. Are Combination Gas and Electric Utilities Multiproduct Natural Monopolies? The Review of Economics and Statistics, vol. 69, no. 3, pp. 392398. State Soil Geographic (STATSG0) Database. 1995. last retrieved in March 2012, http://dbwww.essc.psu.edu/dbtop/doc/statsgo/statsgo_db.pdf. Thompson, H. G. Jr. 1997. Cost Efficiency in Power Procurement and Delivery Service in the Electric Utility Industry. Land Economics, vol. 73, no. 3, pp. 287-296. Verhulst, M. J. J. 1948. The Theory of Production Applied to the French Gas Industry. Econometrica, vol. 16, no. 4, pp. 295-308. Weiss, L. W. 1975. Antitrust in the Electric Power Industry. In Phillips, A. (Ed) Studies in the Regulation of Economic Activity: Promoting Competition in Regulated Markets, The Brookings Institution, Washington D.C., pp. 135-173. Wells, F. J. 1977. The Effects of Customer Density on Electrical Distribution Costs. In Downing P. B. (Ed). Local Service Pricing Policies and Their Effect on Urban Spatial Structure, University of British Columbia Press, Vancouver, pp. 87-118. Yatchew, A. 2000. Scale Economies in Electricity Distribution: A Semiparametric Analysis. Journal of Applied Econometrics, vol. 15, pp. 187-210.

231

Yu, W., T. Jamasb, M. Pollitt. 2009. Does Weather Explain Cost and Quality Performance? An analysis of UK electricity distribution companies. Energy Policy, vol. 37, pp. 4177-4188. Zastrow, O. W. 1967. Underground Corrosion and Electrical Grounding. IEEE Transactions on Industry and General Applications, vol. IGA-3, no.3, pp. 237243.

232

APPENDIX A MODELS OF ELECTRICITY AND NATURAL GAS CUSTOMERS AND SALES

Electricity and natural gas sales and numbers of customers are inputs to investment cost models. Therefore, a complete customers and sales database may generate better estimations of cost functions. Census data and geographic data are available for all 1619 tax districts (1614 tax districts and New York City’s five boroughs). However, actual electricity and natural gas customers and sales data are limited to a smaller number of tax districts. There are 257 observations for actual electricity customers and sales and 210 observations for natural gas customers and sales. There are 206 tax districts where companies in the database serve both electricity and natural gas. However, in 20 of those tax districts, data are available only on natural gas sales. Finally, there are 61 tax districts where they serve electricity exclusively. This chapter has two purposes: (1) expand the existing sales and customers database, and (2) to model market behavior using actual market data.

233

A.1 Introduction

Customers and sales are highly dependent on demographic and housing characteristics, and on economic activities, and therefore could be modeled using demographic and housing characteristics. Missing data for customers and sales could then be estimated using census and geographic data. For example, the total number of residential electric customers in a tax district may be expected to be related to the total number of households and residential land use, while total residential electric sales are related to the number of electricity customers, population density, housing equipment that uses electricity, price of electricity, etc. Variables that affect the number of commercialindustrial customers and sales may include economic census variables, such as numbers of establishments and employees. However, such census data is only available for tax districts with a population over 2,500. Prices are not available in tax districts without market (i.e., utility) data. There are 56 tax districts with manufacturing data, and 110 tax districts with retail, service and wholesale data among the tax districts with electricity sales and customers data. These numbers further drop for natural gas: 45 observations for manufacturing and 90 observations for the other three industries. The price variables are not given in the data directly, but are approximated as the ratios of revenues over sales for both electricity and natural gas, in the residential and commercial-industrial sectors. Limitations in the economic censuses and in the price variables prevent models containing these variables to be used for estimating missing values. However, these data can be used to estimate separate models with smaller sample

234

sizes, where the goal is to explain. Therefore, two sets of electricity and natural gas models are considered separately: estimation models and explanation models. While modeling the numbers of customers and sales, linear, log-linear, and loglog functional forms are considered. The log-linear specification is generally inferior to the linear and log-log ones. When seeking the best model specification, models are compared in terms of functional forms or independent variables set. Linear and log-log models with the same set of independent variables are compared using the d-statistic (Rao and Miller, 1971, pp. 107-111; Guldmann, 1983, pp. 308-310). The method is based on the comparison of the residual sums of squares of the linear model (RSS1) and the loglog model (RSS2). However, the residual sums of squares of the two models cannot be compared directly due to the difference in dependent variables (one in linear form, the other in log form). Therefore, the residual sum of squares of the linear model (RSS1) must be adjusted (RSS1*). If the dependent variable has been standardized by dividing it by its geometric mean, then the two residual sums of squares are comparable (Rao and Miller, 1971, p. 109). To have comparable the residual sums of squares, RSS1 is multiplied by the square of the inverse of the geometric mean of the dependent variable, producing RSS1* for the linear model. The computed d statistic follows the chi-square distribution (Rao and Miller, 1971, p.111; Guldmann, 1983, p. 309):

|

|

(A.1)

The null hypothesis, H0, states that two functions are empirically equivalent if the computed d does not exceed the critical value, dc. This statistics has a chi-square 235

distribution in case of equivalency of the two forms, with one degree of freedom. When the computed d value exceeds the critical dc value, the null hypothesis H0 of equivalence is rejected, hence the two functional forms are not equivalent. The model with the lower RSS is superior to the other model. The F-test is used to analyze the effect of an additional variable. In such models the functional forms are the same and the independent variables are the same, except for the additional ones. The model with less variables is called the constrained model, since it is assumed that the coefficients of the additional variables are equal to zero (

. The

model with more variables is called the unconstrained model, and the F-statistics is:

(A.2)

where

= Residual sums of squares of the constrained model = Residual sums of squares of the unconstrained model M = Number of restrictions (i.e. the number of additional independent variables) N = Number of observations K = Number of independent variables in the unrestricted model

The null hypothesis, H0 , states that additional variables have zero effect, making the two models statistically equivalent. If the computed F value is greater than the critical F (with M degrees of freedom in the numerator and N-K-1 degrees of freedom in the denominator), then the null hypothesis is rejected, concluding that the additional variables have a statistically significant effect on the model. 236

A.2 Estimation Models

Estimation models will be used to extend sales and customers data to all 1619 tax districts. Restricted variables, such as economic censuses and prices, are not used in these estimation models. These models cover the residential and commercial-industrial sectors for electricity and natural gas. Electricity has also a lighting sector.

A.2.1 Electricity Models

A.2.1.1 Modeling the Number of Residential Electricity Customers

The total number of electricity customers (

) is a function of the number of

households (HH) and the share of residential land use (LU_PRES). Almost all households are connected to the grid as electricity customers, and therefore HH is directly related to ; LU_PRES, on the other hand, may capture the variations that are not explained by HH. The best model is:

ln

= 1.03 + 0.878 lnHH + 0.104 ln LU_PRES (4.68) (31.3) (2.29) R2=0.797,

F=490.04,

N=255

(A.3)

Residential land use is informative about residential electricity customers who are not considered as single customers, such as assisted living houses, dormitories, or some

237

residential estates where electricity is included in the rent and the management office is the entity which appears as the only customer. The logarithmic model produces a better fit than the linear one according to the dstatistic. The residual sums of squares are

=1010 and

=131.05, and the number

of observations 255. The computed d value is 113.26, which is greater than the critical d (d=3.841, α=0.05). The distributions of the variables, which are positively skewed (Figure A.1), also support the superiority of the logarithmic function. As a result, a 1 % increase in the total number of households in a tax district would increase the total number of residential electricity customers by 0.87 %, and a 1 % increase in the share of residential land-use would increase this number by 0.1 %.

Figure A.1 Distribution of NRE and HH

238

A.2.1.2 Modeling Residential Electricity Sales

Residential electricity sales (SRE) are a function of the number of customers, major housing equipment using electricity, density, and the average house size. The number of customers is the most important variable that has a direct effect on electricity sales. House heating and cooling, cooking, and water heating are the major housing activities that use electricity. House cooling is related to AC. The house heating, cooking, and water heating variables are highly correlated (Table A.1). House cooking is the variable that appears to be the most significant. In fact, house heating and water heating generally use natural gas, while electricity is preferred for cooking stoves. The shares of houses with AC (AC_P) and using electricity in house cooking (H_PCOOKE) are used. Density and the average house size are also expected to have an effect on sales. The density variable is the ratio of the number of houses over the residential land use area (DENSHOUS), and the average house size is the ratio of the total room numbers over the number of houses (ROOMAVG).

H_PHEATE H_PCOOKE W_PHEATE

Pearson Correlation Sig. (2-tailed) N Pearson Correlation Sig. (2-tailed) N Pearson Correlation Sig. (2-tailed) N

H_PHEATE 1 257 .793** 0 257 .939** 0 257

H_PCOOKE .793** 0 257 1 257 .783** 0 257

W_PHEATE .939** 0 257 .783** 0 257 1 257

Table A.1 Correlations among the Houses Using Electricity for Space Heating (HHEATE), Cooking (HCOOKE), and Water Heating (WHEATE) 239

As with the NRE model, the logarithmic form yields a better fit than the linear form for the SRE models (

=1320, and

=47.1, N= 250). The computed d value is

180.9, greater than critical d. SRE has a positively skewed distribution (Figure A.2), which also points to the logarithmic form.

Figure A.2 Distribution of SRE

The best model is:

ln SRE = 8.99 + 0.942 lnNRE + 0.163 lnAC_P + 0.145 lnH_PCOOKE (20.68) (47.77) (3.01) (3.1) – 0.088lnDENSHOUS + 0.658lnROOMAVG (-2.28) (4.65) R2=0.915, 240

F=526.54,

N=250

(A.4)

The model shows that a 1% increase in the number of customers increases sales by 0.94%. A 1% increases in the shares of houses with AC and the shares of houses using electricity in house cooking increase sales by 0.16% and 0.14%, respectively. Density, on the other hand, has a negative effect: a 1% increase in house density decreases sales by 0.08%. The negative relationship between density and sales can be explained by the advantages of compact urban form. A higher density provides protection from unfavorable weather conditions, and provides more insulation. The average number of rooms is directly related to space-heating needs, and a 1% increase leads to an increase in sales of 0.65%. Also, electricity is needed to light larger houses, which tend to have more electricity-dependent appliances.

A.2.1.3 Modeling the Number of Commercial-Industrial Electricity Customers

Data is available for the Commercial and Industrial sectors separately only for Central Hudson Company. These sectors are combined in the other three companies. Customer and sales observations are summed up for Central Hudson to increase the sample size. The number of Commercial-Industrial (C-I) customers is expected to be the function of the number of economic establishments and of the C-I land use area. However, due to the limitations in the economic censuses, establishment data cannot be used in estimation models. In order to approximate the numbers of establishments, population is used instead. Populated areas tend to have more economic activities, particularly in the retail and service sectors. C-I land use captures the variations that are

241

not explained by the population variable. The logarithmic form is superior to the linear one:

=2021.25,

=199.58, and dc = 111.1. The best model is:

lnNCIE = -1.44 + 0.799 lnPOP + 0.076 lnLUCI (-2.3) (12.26) (2.2) R2=0.696,

F=286.7,

N=253

(A.5)

A 1% increase in population leads to an increase in the number of C-I customers by 0.79%, and a 1% increase in the C-I land use leads to an increase of 0.076%.

A.2.1.4 Modeling Commercial-Industrial Electricity Sales

The number of C-I customers (NCIE) is the most important variable with an effect on sales. More customers increase sales. Electricity sales in the C-I sector are expected to be related to the size of establishments. Larger establishments tend to use more electricity. Total employment could be used to approximate establishment size, however models with employment data are geographically restricted in the economic censuses. The C-I land use area is used instead to approximate the size of the economic activity in a specific area. The best model is:

lnSCIE = 10.41 + 1.11 ln NCIE + 0.096 lnLUCI (33.71) (23.03) (3.4) R2=0.843,

242

F=672.02,

N=253

(A.6)

A 1% increase in the number of C-I customers increases sales by 1.11 percent, and a 1% increase in C-I land use area increases sales by 0.096 percent.

A.2.1.5 Modeling Lighting Customers and Sales

Modeling lighting customers (NL) and sales (SL) is different from modeling the residential and commercial-industrial sectors. It is difficult to clearly delineate who the lighting customers are. Street lighting service is usually provided by local governments and transportation departments. For instance, New York City provides service and maintenance for street lighting within its boroughs. Local street lights are usually paid for by the city or county out of local property taxes. Interstates lights are paid by the state. Therefore, the customer structure is complicated and is hard to define. Given these considerations, lighting customers are modeled as a function of the number of houses (HOUS), the median house value (VAL), and the average street segment length (SEG). The number of houses and median house value are expected to have a positive effect, while the average street segment length (total street length/total number of junctions) is expected to have a negative effect on NL. Street segment length can be considered as a measure of density, with shorter segments implying smaller sub-divisions and higher densities. Compact urban forms would call for fewer customers. The log-log models yield a better fit than the linear ones, due to the non-normal distribution of data (Figure A.3).

243

Figure A.3 Distribution of NL and SL

The best model is:

ln NL = -9.53 + 0.37 lnHOUS + 0.639 lnVAL – 0.526 lnSEG (-8.81) (6.00) (6.48) (-2.48) R2=0.517,

F=84.97,

N=241

(A.7)

Lighting sales can be considered a function of the built-up area, street structure, and density, as well as NL. Built-up area (ABLTP) is the total area of all urban-related land uses, such as residential, commercial, industrial, etc. within the tax district. Density (DENSA) is the area’s population density (population over land area). Street structure is approximated by the number of intersections (INT), since an increasing number of intersections is likely to result in more lighting sales. The price variable, computed as the ratio of revenues over sales (RL/SL) is not statistically significant. The best model is:

244

ln SL = 3.98 + 0.238 ln NL + 0.374 ln ABLTP + 0.700 lnDENSA + 0.459 lnINT (4.61) (3.43) (2.59) (11.39) (2.94) R2=0.657,

F=113.93,

N=241

(A.8)

A.2.2 Natural Gas Models

A.2.2.1 Modeling the Number of Residential Natural Gas Customers

The number of residential natural gas customers (NRG) is a function of population, wealth, density, residential land use area, and type of fuel used in house equipment (heating, cooking and/or water heating). The population variable is the number of households (HH), and it is expected that an increasing number of households would increase NRG. Similarly, residential land use is expected to represent other variables, asides from the number of households, such as apartment complexes, the complex is the customer instead of the individual household. Preferences of customers for natural gas can be explained by the type of fuel used in house equipment (HHEATG, HCOOKG, and WHEATG). The log-log models are superior to the linear ones, using the d-statistics. This is also related to the non-normal distribution of customer and sales data (Figure A.4).

245

Figure A.4 Distribution of NRG and SRG

In contrast to electricity, households do not all need to be natural gas customers. It is basically a choice related to the advantages and availability of natural gas. Natural gas house heating is the most significant equipment variable. The best model is:

lnNRG = 0.09 + 0.352 lnHH + 0.675 lnHHEATG + 0.165 lnLU_PRES (0.24) (3.93) (9.93) (2.18) R2=0.764,

F=215.78,

N=203

(A.9)

A 1% increase in the number of households increases NRG by 0.35%. A 1% increase in the number of houses using utility gas for space heating increases the number of customers by 0.67%. The share of residential land captures effects not explained by the other variables, such as the effect of communities like dormitories, as discussed in the electricity models.

246

A.2.2.2 Modeling Residential Natural Gas Sales

Natural gas sales (SRG) are measured in Mcf units (1000 cubic feet). Sales can be considered as a function of the number of customers (NRG) and population density (DENSA). The shares of houses using natural gas in heating, cooking, and water heating are also important, but, due to their correlations, only the house heating share appears to be significant. The size of the house is also expected to be important. Size is approximated by the average room variable (ROOMAVG). The number of stories is also important when house heating is considered. The best model is:

lnSRG = 4.42 + 0.776 lnNRG + 0.56 lnH_PHEATG + 1.181 lnROOMAVG (6.47) (18.17) (11.12) (5.75) – 0.146 lnDENSA + 0.195 lnSTO1to3 (-3.21) (3.90) R2=0.938,

F=553.85

(A.10)

A 1% increase in the number of customers, in the share of houses using gas for space heating, and in the average number of room leads to increases in residential sales by 0.77%, 0.56%, and 1.18%, respectively. Density has a negative effect as expected. Households need less energy for house heating in denser areas, due to the urban heat island effect and the protection from heavy winds. A 1% density increase decreases utility gas sales by 0.14%. At the micro scale, the number of stories is also important in terms of insulation. Low-rise single-family houses need more energy to heat the house, while in an apartment households benefit from the insulation from their downstairs and 247

upstairs neighbors. The STO1to3 variable is the number of houses with 1 to 3 stories, and it has a positive effect on sales: a 1% increase in the total number of low-rise houses increases sales by 0.19%.

A.2.2.3 Modeling Commercial-Industry Natural Gas Customers

C-I gas sales are estimated using C-I land use area and population density. C-I land use area provides information about the concentration of firms while denser areas are expected to attract economic activities. The best model is:

lnNCIG = -0.44 + 0.821 lnLUCI + 0.7 lnDENSA (-0.6) (18.32) (8.97) R2=0.692,

F=201.1,

N=181

(A.11)

A 1% increase in C-I land use and in population density increases the number of C-I natural gas customers by 0.82% and 0.70%, respectively.

A.2.2.4 Modeling Commercial-Industrial Natural Gas Sales

Natural gas sales in the C-I sector are directly related to the number of C-I customers and the C-I land use area. The land use variable provides information about the intensity of economic activity in that area. The best model is:

248

lnSCIG = 7.45 + 0.831lnNCIG + 0.296lnLUCI (18.5) (11.91) (3.93) R2=0.740,

F=254.22,

N=181

(A.12)

The regression results show that a 1% increase in the number of customers increases sales by 0.83%, and a 1% increase C-I land use area increases sales by 0.29%.

A.3 Explanation Models

Explanation models include a wider range of variables, such as economic census variables and prices. Therefore, they are expected to yield better fits than the estimation models. The price variable is not expected to have a significant effect on the number of residential customers, but is expected to influence the amount of electricity used. Explanatory models are therefore not developed here for residential electricity and natural gas customers.

A.3.1 Electricity Models

A.3.1.1 Modeling Residential Electricity Sales

The price of residential electricity is expected to have a significant effect on sales. The price has a negative effect, as expected. A 1% increase in the residential electricity price decreases sales by 0.56%. The other variables are quite similar to those used in the residential electricity estimation model (Eq. A.4), except for the absence of the house 249

cooking variable, which turns out to be insignificant with the addition of the electricity price (PRE). The addition of the price variable improves the model fit, which points to the importance of price in electricity usage. The best model is:

lnSRE = 8.27 + 0.922 lnNRE + 0.245 lnAC_P – 0.149 lnDENSHOUS (17.9) (4.6) (4.37) (-4.64) + 0.536lnROOMAVG – 0.568ln PRE (3.79) (-4.74) R2=0.919,

F=555.64

N=250

(A.13)

A.3.1.2 Modeling Commercial-Industrial Electricity Customers

The number of establishments is considered while modeling C-I customers. First, models are estimated with data for the disaggregated sectors (manufacturing, retail, service, and wholesale). Then, the four sectors are aggregated and models are estimated using aggregated data. Zero values are assigned to observations without economic census information. There is a significant correlation among the four sectors in the disaggregated models (Table A.2), which makes it impossible to put them all in the same model. The logarithmic form is superior to the linear one. The economic census variables are converted using the Box-Cox transformation due to the number of zero observations. Here, the number of manufacturing establishments (MAEST) is a Box-Cox converted variable, with a parameter of 0.001. The most important variable appears to be manufacturing. Population and land use variables are also significant. The best model is:

250

lnNCIE = -0.39 + 0.001 MAEST + 0.737 lnPOP + 0.076 lnLUCI (-0.59) (4.20) (11.38) (2.28) R2=0.716,

MAEST MAEST

Pearson Correlation

1

Sig. (2-tailed) N REEST

SEEST

WSEST

Pearson Correlation

257 .752

F=209.74,

N=253

REEST .752**

SEEST .777**

WSEST .789**

.000

.000

.000

257

257

257

1

**

.952**

.000

.000

**

.960

Sig. (2-tailed)

.000

N

257

257

257

257

.777**

.960**

1

.976**

Sig. (2-tailed)

.000

.000

N

257

257

257

257

.789**

.952**

.976**

1

Sig. (2-tailed)

.000

.000

.000

N

257

257

257

Pearson Correlation

Pearson Correlation

(A.14)

.000

257

**. Correlation is significant at the 0.01 level (2-tailed).

Table A.2 Correlations among the Number of Establishments: Manufacturing (MAEST), Retail (REEST), Services (SEEST), Wholesale (WSEST)

A 1% increase in the number of manufacturing establishments increases C-I customers by 0.001% at the sample mean (ε=0.001*980.001). A 1% increase in population increases C-I customers by 0.73%, and a 1% increase in C-I land use area increases C-I customers by 0.076%. The aggregated models do not yield as good a fit as the disaggregated models. The price of C-I electricity is not significant in any model. 251

A.3.1.3 Modeling Commercial-Industrial Electricity Sales

Both disaggregated and aggregated explanation models were considered. Employments in manufacturing, retail, service, and wholesale sectors are highly correlated (Table A.3), and therefore cannot all be simultaneously entered into the same model. As for the NCIE models, the only sector that appears to be significant is manufacturing. However, when the price variable enters the model, the employment variable turns out to be insignificant, and the price variable improves the model more than the other variables. In contrast to other sales models, a meteorological variable, the annual total cooling degree days (CDDSUM), appears to be significant in the C-I electricity sales model.

MAEMP MAEMP

REEMP

SEEMP

WEEMP

Pearson Correlation Sig. (2-tailed) N Pearson Correlation Sig. (2-tailed) N Pearson Correlation Sig. (2-tailed) N Pearson Correlation

1 257 .715** .000 257 .801** .000 257 .795**

REEMP .715** .000 257 1 257 .943** .000 257 .954**

SEEMP .801** .000 257 .943** .000 257 1 257 .966**

Sig. (2-tailed)

.000

.000

.000

N

257

257

257

WSEMP .795** .000 257 .954** .000 257 .966** .000 257 1 257

**. Correlation is significant at the 0.01 level (2-tailed).

Table A.3 Correlations among the Number of Employees: Manufacturing (MAEMP), Retail (REEMP), Services (SEEMP), Wholesale (WSEMP) 252

The best model is: lnSCIE = -4.7 + 1.01 lnNCIE + 0.064 lnLU_PCI – 2.73 lnPCE + 1.26 lnCDDSUM (-2.24) (28.78) (2.70) (-11.76) (4.97) R2=0.899,

F=553.2,

N=253

(A.15)

The addition of the C-I price variable increases model fit in a very significant way, and a 1% increase in price decreases C-I sales by 2.73%. The share of C-I land use is a proxy for economic activity, and a 1% increase in LU_PCI increases sales by 0.064%. A higher need for cooling results in larger electricity sales: a 1% increase in total cooling degree days increases sales by 1.26%.

A.3.2 Natural Gas Models

A.3.2.1 Modeling Residential Natural Gas Sales

Using natural gas as the fuel for house appliances (heating, cooking, etc.) is a choice, and price is expected to be an important variable affecting customer choice and behavior. Equation (A.16) shows that the price variable has a statistically significant effect, with the expected negative sign. A 1% increase in residential utility gas price leads to a 3.92% decrease in sales. The number of residential customers and the share of residential land use have also statistically significant effects on sales, with 1% increases resulting in a 0.89% and a 0.3% increase in sales, respectively. The average number of rooms and the number of low-rise houses capture the impact of urban form on sales. 253

Spacious, low-storied residential areas tend to spend more on natural gas, particularly for house heating. The model explains about 96% of the variations in residential natural gas sales. The best model is:

lnSR = 9.18 + 0.897 lnNR + 0.305 lnLU_PRES + 0.981 lnROOMAVG (15.69) (31.62) (7.45) (5.83) – 3.929 lnPRG + 0.093 ln STO1to3 (-16.98) (2.41) R2=0.959,

F=868.98,

N=187

(A.16)

A.3.2.2 Modeling Commercial-Industrial Natural Gas Customers

In the explanation models, the economic census and price variables are considered. However, economic census variables in terms of numbers of establishments, both disaggregated and aggregated, do not improve the model fit. The price variable, on the other hand, improves the model in a statistically significant way. A 1% increase in price decreases the number of customers by 1.06%. The wealth of the area provides information about economic activity, particularly in the retail and service sectors. Therefore, median house value of the tax district is added to the model, and a 1% increase in this variable increases the number of C-I natural gas customers by 0.24%. The best model is:

lnNCIG = -2.14 + 0.688 lnLUCI + 0.729 lnDENSA – 1.065 lnPCIG + 0.24lnVAL (-1.35) (10.18) (9.35) (-2.34) (2.21) R2=0.706, 254

F=106.25,

N=181

(A.17)

A.3.2.3 Modeling Commercial-Industrial Natural Gas Sales

The price variable is important for firms using natural gas. A 1% increase in C-I gas prices decreases sales by 4.14%. The effects of C-I land use and the number of C-I customers are smaller than the price effect.

lnSCIG = 13.7 + 0.77 lnNCIG + 0.233 lnLUCI – 4.14 lnPCIG (19.9) (13.79) (3.87) (-10.26) R2=0.836,

F=303.29,

N=181

(A.18)

A.4 Concluding Remarks

The number of residential and C-I customers and sales of electricity and natural gas are estimated using census data and geographic variables. Price is another influential factor. However, price data is not available for districts where market data must be estimated. Therefore, two types of models have been considered: one type with available data throughout the state, the other with a price variable approximated by the ratio of revenues over sales in each sector. The first type of models will be tested to possibly expand the sample size for the estimations of electricity and natural gas investment cost models (Chapter 5). The second type of model is explanatory, showing the effect of prices. Density and land use are the site-specific variables that are common in most models. When house appliances are considered, air conditioning and house cooking are the significant variables in residential electricity sales, while house heating is significant 255

in residential natural gas sales. The average number of rooms, as an indicator of house size, is also significant in residential gas sales. The number of C-I customers and sales models have limitations due to the unavailability of economic census data in areas with a population less than 2500. Models with price variable have better fits, as expected, and price is approximated by the ratio of revenues over sales.

256

APPENDIX B ECONOMIES OF SCALE IN AGGREGATE CAPITAL INVESTMENT B.1 Economies of Scale in Aggregate Electricity Capital Investment

Economies of scale are measured by:

(B.1) where

θ=0.093

The Box-Cox Model is:

( (

) )

( (

) )

Therefore:

257

( (

) )

(

)

(B.2)

[

(

(

)

(

)

)

(B.3)

]

or:

[

(B.4)

]

For any given values for the output variables (NRE, SRE, SCIE),

is at its

maximum when the area density, DENSA, and the electricity load factor, LFE, are at their maximum, and soil corrosivity, SOILCORR, and the number of street intersections, INT, are at their minimum, with:

DENSAmax = 103,058 SOILCORRmin = 2.78 INTmin = 13 LFEmax = 0.63

For the above values, it follows that:

(B.5)

258

(B.6) Express all outputs as functions of k: ̅ ̅

̅

̅

where: ̅ ̅ ̅

(minimum value)

̅

It follows that:

(B.7)

where:

(B.8) (B.9)

The next step is to determine the output level k for which constant returns to scale are achieved:

259

or

(B.10)

̅ ̅

̅

̅

(B.11)

Finally: (B.12)

The solution value for k is unrealistically large, which implies that constant returns to scale can never be achieved over the observed output range, even when the site-specific and company-specific variables are assigned the most favorable values.

B.2 Economies of Scale in Aggregate Natural Gas Capital Investment

Economies of scale are measured by:

(B.13)

where 260

θ=0.206 and

Define:

H=

(B.14)

Then:

[

(B.15)

]

It follows that:

(B.16)

Which implies:

(B.17) 261

or:

(B.18)

There are 3 tax districts where diseconomies of scale are observed in natural gas distribution6. When H (Eq. B.14) is computed with the outputs of these communities, and when using the corresponding ABLTP and WPEG values, the trade-offs between SOILWORK and INT can be computed, as illustrated on Figure B.1.

(a) Figure B.1 Constant Returns to Scale Tradeoff Curve ( ) for the Three Communities where : (a). Havestraw Village, (b) Wappingers Falls Village, (c) Grandview on Hudson

6

See Section 5.1.3.2, Table 5.7.

262

Figure B.1 continued

(b)

(c)

263

APPENDIX C SCOPE SCORES FOR DIFFERENT LEVELS OF SITE-SPECIFIC VARIABLES AND DIFFERENT OUTPUT COMBINATIONS C.1 Scope Score Tables for Aggregate Multi-Utility Costs

ELECTRICITY (kwh)

NATURAL GAS (mcf)

83 346,061 692,039 1,038,018 1,383,996 1,729,974 2,075,952 2,421,930 2,767,908 3,113,865 0.14 0.709

0.309

0.279

0.262

0.251

0.242

0.235

0.230

0.225

0.221

1,448 0.499

0.261

0.239

0.227

0.218

0.212

0.207

0.202

0.198

0.195

2,896 0.473

0.254

0.233

0.222

0.213

0.207

0.202

0.198

0.194

0.191

4,345 0.457

0.249

0.229

0.218

0.210

0.204

0.199

0.195

0.191

0.188

5,793 0.446

0.246

0.227

0.215

0.208

0.202

0.197

0.193

0.189

0.186

7,241 0.438

0.243

0.224

0.213

0.206

0.200

0.195

0.191

0.188

0.185

8,690 0.430

0.241

0.222

0.212

0.204

0.198

0.194

0.190

0.187

0.184

10,138 0.424

0.239

0.221

0.210

0.203

0.197

0.193

0.189

0.185

0.182

11,586 0.419

0.237

0.219

0.209

0.202

0.196

0.191

0.188

0.184

0.181

13,035 0.414

0.236

0.218

0.208

0.201

0.195

0.190

0.187

0.183

0.181

Table C.1 Scope Scores for the Minimum Share of 40+ Year Old Houses (AGE_P40=0.003)

264

ELECTRICITY (kwh)

NATURAL GAS (mcf)

83 346,061 692,039 1,038,018 1,383,996 1,729,974 2,075,952 2,421,930 2,767,908 3,113,865 0.14 0.786

0.403

0.369

0.349

0.336

0.325

0.317

0.310

0.304

0.299

1,448 0.600

0.348

0.322

0.307

0.296

0.288

0.282

0.276

0.272

0.268

2,896 0.575

0.339

0.315

0.300

0.290

0.282

0.276

0.271

0.266

0.262

4,345 0.560

0.334

0.310

0.296

0.286

0.279

0.273

0.268

0.263

0.259

5,793 0.549

0.330

0.306

0.293

0.283

0.276

0.270

0.265

0.261

0.257

7,241 0.540

0.326

0.304

0.290

0.281

0.274

0.268

0.263

0.259

0.255

8,690 0.533

0.324

0.301

0.288

0.279

0.272

0.266

0.261

0.257

0.253

10,138 0.526

0.322

0.299

0.286

0.277

0.270

0.264

0.260

0.255

0.252

11,586 0.521

0.319

0.298

0.285

0.276

0.269

0.263

0.258

0.254

0.251

13,035 0.516

0.318

0.296

0.283

0.274

0.268

0.262

0.257

0.253

0.249

Table C.2 Scope Scores for the Maximum Share of 40+ Year Old Houses (AGE_P40=0.90)

ELECTRICITY (kwh)

NATURAL GAS (mcf)

83 346,061 692,039 1,038,018 1,383,996 1,729,974 2,075,952 2,421,930 2,767,908 3,113,865 0.14 0.471

0.141

0.124

0.115

0.109

0.105

0.101

0.098

0.096

0.094

1,448 0.267

0.115

0.103

0.097

0.093

0.090

0.087

0.085

0.083

0.082

2,896 0.247

0.111

0.100

0.094

0.090

0.087

0.085

0.083

0.081

0.080

4,345 0.236

0.108

0.098

0.093

0.089

0.086

0.083

0.082

0.080

0.078

5,793 0.228

0.107

0.097

0.091

0.088

0.085

0.082

0.081

0.079

0.077

7,241 0.222

0.105

0.096

0.090

0.087

0.084

0.082

0.080

0.078

0.077

8,690 0.217

0.104

0.095

0.090

0.086

0.083

0.081

0.079

0.078

0.076

10,138 0.213

0.103

0.094

0.089

0.085

0.083

0.080

0.079

0.077

0.076

11,586 0.209

0.102

0.093

0.088

0.085

0.082

0.080

0.078

0.076

0.075

13,035 0.206

0.102

0.093

0.088

0.084

0.082

0.079

0.078

0.076

0.075

Table C.3 Scope Scores for the Minimum Number of Street Intersections (INT=13)

265

ELECTRICITY (kwh)

NATURAL GAS (mcf)

83 346,061 692,039 1,038,018 1,383,996 1,729,974 2,075,952 2,421,930 2,767,908 3,113,865 0.14 0.868

0.547

0.511

0.490

0.475

0.463

0.454

0.446

0.439

0.433

1,448 0.729

0.488

0.459

0.442

0.430

0.420

0.413

0.406

0.400

0.395

2,896 0.708

0.479

0.451

0.434

0.423

0.413

0.406

0.399

0.394

0.389

4,345 0.695

0.473

0.446

0.429

0.418

0.409

0.401

0.395

0.390

0.385

5,793 0.685

0.468

0.442

0.426

0.414

0.406

0.398

0.392

0.387

0.382

7,241 0.677

0.465

0.438

0.423

0.412

0.403

0.396

0.390

0.384

0.380

8,690 0.671

0.462

0.436

0.420

0.409

0.401

0.393

0.387

0.382

0.378

10,138 0.665

0.459

0.433

0.418

0.407

0.399

0.392

0.386

0.380

0.376

11,586 0.661

0.457

0.431

0.416

0.405

0.397

0.390

0.384

0.379

0.374

13,035 0.656

0.454

0.429

0.414

0.404

0.395

0.388

0.382

0.377

0.373

Table C.4 Scope Scores for the Maximum Number of Street Intersections (INT=15,114)

C.2 Scope Score Tables for Underground Multi-Utility Costs

ELECTRICITY (kwh)

NATURAL GAS (mcf)

304 346,323 692,342 1,038,360 1,384,379 1,730,398 2,076,417 2,422,436 2,768,455 3,113,865 0.14 0.411

0.201

0.181

0.170

0.162

0.156

0.151

0.147

0.144

0.141

1,448 0.318

0.176

0.160

0.151

0.145

0.140

0.137

0.133

0.131

0.128

2,896 0.305

0.172

0.157

0.148

0.142

0.138

0.134

0.131

0.128

0.126

4,345 0.296

0.169

0.155

0.146

0.140

0.136

0.132

0.129

0.127

0.124

5,793 0.290

0.167

0.153

0.145

0.139

0.135

0.131

0.128

0.125

0.123

7,241 0.285

0.165

0.151

0.143

0.138

0.133

0.130

0.127

0.125

0.122

8,690 0.281

0.164

0.150

0.142

0.137

0.133

0.129

0.126

0.124

0.122

10,138 0.277

0.163

0.149

0.141

0.136

0.132

0.128

0.126

0.123

0.121

11,586 0.274

0.162

0.148

0.141

0.135

0.131

0.128

0.125

0.122

0.120

13,035 0.271

0.161

0.148

0.140

0.135

0.130

0.127

0.124

0.122

0.120

Table C.5 Scope Scores for the Minimum Built-up Area (ABLTP=0.04)

266

ELECTRICITY (kwh)

NATURAL GAS (mcf)

304 346,323 692,342 1,038,360 1,384,379 1,730,398 2,076,417 2,422,436 2,768,455 3,113,865 0.14 0.666

0.418

0.387

0.368

0.355

0.345

0.337

0.330

0.324

0.319

1,448 0.571

0.379

0.353

0.337

0.326

0.318

0.311

0.305

0.300

0.295

2,896 0.555

0.372

0.347

0.332

0.321

0.313

0.306

0.300

0.295

0.291

4,345 0.545

0.367

0.343

0.328

0.318

0.310

0.303

0.297

0.293

0.288

5,793 0.538

0.364

0.340

0.325

0.315

0.307

0.301

0.295

0.290

0.286

7,241 0.532

0.361

0.337

0.323

0.313

0.305

0.299

0.293

0.289

0.285

8,690 0.527

0.359

0.335

0.321

0.311

0.304

0.297

0.292

0.287

0.283

10,138 0.522

0.357

0.333

0.320

0.310

0.302

0.296

0.290

0.286

0.282

11,586 0.519

0.355

0.332

0.318

0.308

0.301

0.295

0.289

0.285

0.281

13,035 0.515

0.353

0.331

0.317

0.307

0.300

0.293

0.288

0.284

0.280

Table C.6 Scope Scores for the Maximum Built-up Area (ABLTP=133.85)

ELECTRICITY (kwh)

NATURAL GAS (mcf)

304 346,323 692,342 1,038,360 1,384,379 1,730,398 2,076,417 2,422,436 2,768,455 3,113,865 0.14 0.514

0.276

0.251

0.236

0.226

0.219

0.213

0.207

0.203

0.199

1,448 0.414

0.245

0.225

0.213

0.205

0.198

0.193

0.189

0.185

0.182

2,896 0.399

0.239

0.220

0.209

0.201

0.195

0.190

0.186

0.182

0.179

4,345 0.389

0.236

0.217

0.206

0.198

0.192

0.188

0.184

0.180

0.177

5,793 0.382

0.233

0.215

0.204

0.196

0.191

0.186

0.182

0.179

0.176

7,241 0.376

0.231

0.213

0.202

0.195

0.189

0.185

0.181

0.177

0.174

8,690 0.372

0.229

0.211

0.201

0.194

0.188

0.183

0.180

0.176

0.173

10,138 0.368

0.228

0.210

0.200

0.193

0.187

0.182

0.179

0.175

0.173

11,586 0.364

0.226

0.209

0.199

0.192

0.186

0.182

0.178

0.175

0.172

13,035 0.361

0.225

0.208

0.198

0.191

0.185

0.181

0.177

0.174

0.171

Table C.7 Scope Scores for the Minimum Soil Corrosivity (SOILCORR=0.03)

267

ELECTRICITY (kwh)

NATURAL GAS (mcf)

304 346,323 692,342 1,038,360 1,384,379 1,730,398 2,076,417 2,422,436 2,768,455 3,113,865 0.14 0.542

0.299

0.273

0.257

0.247

0.238

0.232

0.226

0.222

0.218

1,448 0.442

0.266

0.245

0.232

0.224

0.217

0.211

0.207

0.203

0.199

2,896 0.426

0.260

0.240

0.228

0.219

0.213

0.208

0.203

0.200

0.196

4,345 0.416

0.256

0.237

0.225

0.217

0.210

0.205

0.201

0.197

0.194

5,793 0.409

0.254

0.234

0.223

0.215

0.209

0.204

0.199

0.196

0.192

7,241 0.403

0.251

0.232

0.221

0.213

0.207

0.202

0.198

0.194

0.191

8,690 0.398

0.249

0.231

0.220

0.212

0.206

0.201

0.197

0.193

0.190

10,138 0.394

0.248

0.229

0.218

0.211

0.205

0.200

0.196

0.192

0.189

11,586 0.390

0.246

0.228

0.217

0.210

0.204

0.199

0.195

0.191

0.188

13,035 0.387

0.245

0.227

0.216

0.209

0.203

0.198

0.194

0.191

0.187

Table C.8 Scope Scores for the Maximum Soil Corrosivity (SOILCORR=0.98)

ELECTRICITY (kwh)

NATURAL GAS (mcf)

304 346,323 692,342 1,038,360 1,384,379 1,730,398 2,076,417 2,422,436 2,768,455 3,113,865 0.14 0.673

0.426

0.394

0.375

0.362

0.352

0.344

0.337

0.331

0.325

1,448 0.579

0.386

0.360

0.344

0.333

0.324

0.317

0.311

0.306

0.302

2,896 0.563

0.379

0.353

0.338

0.328

0.319

0.313

0.307

0.302

0.297

4,345 0.553

0.374

0.349

0.335

0.324

0.316

0.309

0.304

0.299

0.295

5,793 0.545

0.371

0.346

0.332

0.322

0.314

0.307

0.301

0.297

0.292

7,241 0.539

0.368

0.344

0.330

0.320

0.312

0.305

0.300

0.295

0.291

8,690 0.534

0.366

0.342

0.328

0.318

0.310

0.304

0.298

0.293

0.289

10,138 0.530

0.364

0.340

0.326

0.316

0.308

0.302

0.297

0.292

0.288

11,586 0.526

0.362

0.339

0.325

0.315

0.307

0.301

0.296

0.291

0.287

13,035 0.523

0.360

0.337

0.323

0.314

0.306

0.300

0.294

0.290

0.286

Table C.9 Scope Scores for the Minimum Share of Buildings with 1 to 3 Stories (STO_P1to3=0.64)

268

ELECTRICITY (kwh)

NATURAL GAS (mcf)

304 346,323 692,342 1,038,360 1,384,379 1,730,398 2,076,417 2,422,436 2,768,455 3,113,865 0.14 0.524

0.284

0.258

0.243

0.233

0.225

0.219

0.214

0.209

0.205

1,448 0.423

0.252

0.231

0.219

0.211

0.204

0.199

0.195

0.191

0.188

2,896 0.408

0.246

0.226

0.215

0.207

0.201

0.196

0.191

0.188

0.185

4,345 0.398

0.242

0.223

0.212

0.204

0.198

0.193

0.189

0.186

0.183

5,793 0.391

0.240

0.221

0.210

0.202

0.196

0.192

0.188

0.184

0.181

7,241 0.385

0.238

0.219

0.208

0.201

0.195

0.190

0.186

0.183

0.180

8,690 0.380

0.236

0.218

0.207

0.200

0.194

0.189

0.185

0.182

0.179

10,138 0.376

0.234

0.216

0.206

0.198

0.193

0.188

0.184

0.181

0.178

11,586 0.373

0.233

0.215

0.205

0.197

0.192

0.187

0.183

0.180

0.177

13,035 0.369

0.231

0.214

0.204

0.196

0.191

0.186

0.183

0.179

0.176

Table C.10 Scope Scores for the Maximum Share of Buildings with 1 to 3 Stories (STO_P1to3=0.99)

ELECTRICITY (kwh)

NATURAL GAS (mcf)

304 346,323 692,342 1,038,360 1,384,379 1,730,398 2,076,417 2,422,436 2,768,455 3,113,865 0.14 0.477

0.247

0.224

0.210

0.201

0.194

0.188

0.184

0.180

0.176

1,448 0.378

0.218

0.199

0.189

0.181

0.175

0.171

0.167

0.164

0.161

2,896 0.363

0.213

0.195

0.185

0.178

0.172

0.168

0.164

0.161

0.158

4,345 0.354

0.209

0.192

0.182

0.175

0.170

0.166

0.162

0.159

0.156

5,793 0.347

0.207

0.190

0.181

0.174

0.168

0.164

0.161

0.158

0.155

7,241 0.342

0.205

0.189

0.179

0.172

0.167

0.163

0.159

0.156

0.154

8,690 0.337

0.203

0.187

0.178

0.171

0.166

0.162

0.158

0.155

0.153

10,138 0.333

0.202

0.186

0.177

0.170

0.165

0.161

0.158

0.155

0.152

11,586 0.330

0.201

0.185

0.176

0.169

0.164

0.160

0.157

0.154

0.151

13,035 0.327

0.200

0.184

0.175

0.168

0.163

0.159

0.156

0.153

0.151

Table C.11 Scope Scores for the Minimum Average Heating Degree Days (HDD=454)

269

ELECTRICITY (kwh)

NATURAL GAS (mcf)

304 346,323 692,342 1,038,360 1,384,379 1,730,398 2,076,417 2,422,436 2,768,455 3,113,865 0.14 0.641

0.392

0.361

0.343

0.330

0.321

0.313

0.306

0.301

0.296

1,448 0.544

0.353

0.328

0.313

0.303

0.294

0.288

0.282

0.277

0.273

2,896 0.528

0.346

0.322

0.308

0.298

0.290

0.283

0.278

0.273

0.269

4,345 0.518

0.342

0.318

0.304

0.294

0.287

0.280

0.275

0.270

0.266

5,793 0.510

0.339

0.315

0.302

0.292

0.284

0.278

0.273

0.268

0.264

7,241 0.504

0.336

0.313

0.300

0.290

0.282

0.276

0.271

0.267

0.263

8,690 0.499

0.334

0.311

0.298

0.288

0.281

0.275

0.270

0.265

0.261

10,138 0.495

0.332

0.309

0.296

0.287

0.279

0.273

0.268

0.264

0.260

11,586 0.491

0.330

0.308

0.295

0.285

0.278

0.272

0.267

0.263

0.259

13,035 0.488

0.329

0.307

0.294

0.284

0.277

0.271

0.266

0.262

0.258

Table C.11 Scope Scores for the Maximum Average Heating Degree Days (HDD=731)

270