The Pennsylvania State University. The Graduate School DYNAMIC EFFICIENCY MODEL: IN THE U.S. ELECTRICITY INDUSTRY. A Thesis in

Transcription

1 The Pennsylvni Stte University The Grdute School Deprtment of Energy, Environmentl nd Minerl Economics DYNMIC EFFICIENCY MODEL: N NLYSIS OF EFFICIENCY ND DEREGULTION IN THE U.S. ELECTRICITY INDUSTRY Thesis in Energy, Environmentl nd Minerl Economics y Supwt Rungsuriywioon 2003 Supwt Rungsuriywioon Sumitted in Prtil Fulfillment of the Requirements for the Degree of Doctor of Philosophy My 2003

2 We pprove the thesis of Supwt Rungsuriywioon. Dte of Signture Spiro E. Stefnou Professor of griculturl Economics Thesis dvisor Chir of Committee Edwrd C. enice ssistnt Professor of griculturl Economics Timothy. Considine Professor of Nturl Resource Economics dm Z. Rose Professor of Energy, Environmentl nd Regionl Economics Hed of the Deprtment of Energy, Environmentl nd Minerl Economics

3 iii BSTRCT The sttic production efficiency model nd the dynmic dulity model of intertemporl decision ming using prmetric pproch hve een continuously developed ut in seprte directions. The prmetric pproch tes sttisticl noise into ccount, which cn provide ccurte mesures in stochstic environment. In this study the sttic shdow price pproch nd the dynmic dulity model of intertemporl decision ming re integrted to formulte theoreticl nd econometric models of dynmic efficiency in the presence of intertemporl cost minimizing firm ehvior. The dynmic efficiency model mesures the firms inefficiency nd ccounts for lloctive nd technicl inefficiencies of net investment nd vrile inputs. theoreticlly consistent method to mesure the decomposition of dynmic totl fctor productivity growth in the presence of inefficiency is used to construct the dynmic totl fctor productivity growth tht is djusted for devitions from the long-run equilirium within n djustment-cost frmewor. This leds to the specific decomposition of the efficiency gin or loss s source in explining the growth. The dynmic efficiency model is implemented empiriclly using pnel dt set of 72 U.S. mjor investor-owned electric utilities using fossil fuel-fired stem electric power genertion during the time period of 986 to 999. Given qudrtic functionl form to specify vlue function of the dynmic progrmming eqution, the dynmic efficiency model is pplied to estimte the cost inefficiency of electric utilities nd to decompose the cost inefficiency into lloctive nd technicl inefficiencies of net

4 iv investment nd vrile inputs, nd to chrcterize the cost structure of the electric utilities under dynmic djustment. The decomposition of the dynmic totl fctor productivity growth in the presence of inefficiency is estimted s mens to evlute how different electric utilities prticipte in the deregultion of electricity genertion. Finding tht the electric utilities re perfectly techniclly efficient in net investment, the empiricl results of this study re summrized s follows. First, electric utilities with reltively high technicl inefficiency of vrile input demnd in sttes dopting deregultion pln improve the performnce of the utilities. The results indicte tht the difference of lloctive efficiencies of vrile inputs nd net investment y the group of electric utilities ffected y the deregultion pln is not significnt. Second, the estimtes of the input price elsticities indicte the sustitution possiilities mong the inputs. Most electric utilities in this study re over-cpitlized in the production nd the estimtes suggest the supporting evidence of incresing returns to scle in the production of the electricity industry. Finlly, the results indicte tht electric utilities locted within sttes with the deregultion pln hd lower verge nnul totl fctor productivity growth thn those locted outside of these sttes. The lower totl fctor productivity growth of electric utilities locted within sttes with the deregultion pln results from the lower technologicl progress nd comined scle effects ut they hve higher comined efficiency effect thn those locted outside of these sttes. The results suggest tht electric utilities locted within sttes with the deregultion pln nticipted deregultion to greter extent thn those locted outside these sttes.

5 v TBLE OF CONTENTS Pge LIST OF FIGURES...ix LIST OF TBLES...xi CKNOWLEDGEMENTS... xv Chpter : INTRODUCTION..... Sttement of Prolem Motivtion Ojectives Chpter Orgniztion... 8 Chpter 2: LITERTURE REVIEWS Development of Dynmic Fctor Demnds Intertemporl Firm Behvior Mesurement of Productivity nd Efficiency Non-prmetric pproch to the Mesurement of Productivity nd Efficiency Prmetric pproch to the Mesurement of Productivity nd Efficiency Mesurement of Totl Fctor Productivity Growth Decomposition... 28

6 vi 2.4. Empiricl Studies of Mesurement of Efficiency nd Productivity in the Production of U.S. Electricity Industry Concluding Comments Chpter 3: THEORETICL BCKGROUND ON THE ECONOMICS OF EFFICIENCY ND DYNMIC FCTOR DEMNDS Decomposition of Economic Efficiency nd Shdow Cost pproch Defining Efficiency Concepts in Production, Distnce, nd Cost Functions Shdow Cost Model Chrcterizing lloctive nd Technicl Efficiencies Intertemporl Dulity Theory Isoqunt nd Isocost Curves Long-Run Cost Curve Concluding Remrs Chpter 4: SPECIFICTION ND ESTIMTION OF DYNMIC EFFICIENCY MODEL Specifiction of Dynmic Efficiency Model Dynmic Efficiency Model Derivtion Functionl Form Specifiction Estimtion pproch... 66

7 vii 4.3. Dynmic Structure of Production Input Demnd Elsticities Scle Elsticity Cost Elsticity Concluding Remrs Chpter 5: INDUSTRY BCKGROUND ND DISCUSSION OF DT Bcground on the Electricity Industry in the United Sttes Issues in Restructuring the U.S. Electricity Industry Discussion of Dt Concluding Remrs Chpter 6: EMPIRICL RESULTS Estimtion Results of the Bse Cse Model Sensitivity nlysis Hypotheses Testing Estimtes of Input Demnd Elsticities Estimtes of Dynmic Cost Structures Concluding Remrs... 32

8 viii Chpter 7: DECOMPOSITION OF DYNMIC TOTL FCTOR PRODUCTIVITY GROWTH IN THE PRESENCE OF EFFICIENCY Decomposition of Dynmic Productivity Growth in the Presence of Inefficiency Overll Time Period Results Comprison of the Results of the Regulted nd the Deregulted Firms Mesure of Fctor Using Bis in Technologicl Chnge Concluding Remrs Chpter 8: CONCLUSIONS ND FUTURE EXTENSIONS Summry of Empiricl Results Suggestions for the Future Reserch BIBLIOGRPHY PPENDIX : LISTS OF ELECTRIC UTILITIES ND DESCRIPTIVE STTISTICS OF VRIBLES PPENDIX B: ESTIMTION RESULTS OF DYNMIC EFFICIENCY MODEL. 97 PPENDIX C: ESTIMTION RESULTS OF DYNMIC TOTL FCTOR PRODUCTIVITY DECOMPOSITION... 23

9 ix LIST OF FIGURES Figure 3-: The Input-Oriented Mesurement nd Decomposition of Cost Efficiency Figure 3-2: The Shdow Price Isocost... 4 Figure 3-3: The Internl Cost ssocited with djusting Qusi-Fixed Fctors in Terms of Foregone Output Figure 3-4: The Reltionship of the Discounted Long-Run Shdow Cost Function, the Shdow Vlue of Cpitl nd the Cpitl Stocs Figure 3-5: The Isoqunt nd Isocost Line for (x,i)... 5 Figure 5-: Percentge Shres of Electricity Genertion y Technology nd Fuel of the United Sttes in Figure 5-2: Percentge Shres of Totl Generting Cpility nd of Finl Sles y Type of Utility in Figure 5-3: The Plot of the Percentge Chnge of Vlue of Cpitl y Wisconsin Pulic Service (WI) Figure 6-: Sensitivity Plot of Technicl Inefficiency for Net Investment... 6 Figure 6-2: Distriution of the Rtio of the Optiml Cpitl to ctul Cpitl Figure 6-3: The Long-Run Cost Curve for the Stem Electricity Genertion Industry... 3

10 x Figure 7-: Percentge Shres of the Scle-Relted Components of TFP Figure 7-2: Percentge Shres of the Efficiency-Relted Components of TFP Figure 7-3: Firm Distriution of Totl Fctor Productivity Figure 7-4: Plots of Totl Fctor Productivity Growth over Time Figure 7-5: Plots of Totl Fctor Productivity Growth y Group of Firms ffected to Deregultion Pln nd y Clsses of TFP Growth Firms... 6 Figure 7-6: Plots of Totl Fctor Productivity Growth y Group of Firms ffected to Deregultion Pln nd y Clsses of Efficiency Firms... 65

11 xi LIST OF TBLES Tle 5-: ctions nd Roles of Stte nd Federl Regultors... 8 Tle 5-2: Sttus of Stte Electric Industry Restructuring ctivity Tle 5-3: Comprison of Restructuring Policies etween Cliforni nd Pennsylvni Tle 5-4: Units, Dt Level nd Sources of the Dt Tle 5-5: Summry of the Dt of 72 Electric Utilities over the Periods of Tle 5-6: Distriution of Firm Output Tle 5-7: Distriution of Cpitl Vlue Tle 6-: Estimted Structurl Coefficients, 3SLS Estimtion Period 986 to 999 (Full Model) Tle 6-2: Estimted Structurl Coefficients, GMM Estimtion Period 986 to999 (Full Model) Tle 6-3: verge Firm lloctive nd Technicl Efficiencies of Net Investment nd of Vrile Input Demnds, Given τ = Tle 6-4: Distriution of verge Technicl Inefficiency of Vrile Input Demnds... Tle 6-5: Distriution of verge lloctive Inefficiency of Vrile Input Demnds...

12 xii Tle 6-6: Distriution of verge lloctive Inefficiency of Net Investment... 2 Tle 6-7: Reltive Chnge of Inefficiency Prmeters over Time... 2 Tle 6-8: Sensitivity nlysis of Interest Rte, r %... 5 Tle 6-9: Sensitivity nlysis of Technicl Inefficiency for Net Investment t r = 5 %... 5 Tle 6-0: Estimted Structurl Coefficients, GMM Estimtion Period 986 to 999 (Restricted model)... 9 Tle 6-: Estimted Structurl Coefficients, GMM Estimtion Period 986 to 996 (Full Model) Tle 6-2: Hypotheses Testing of the Dynmic Structure... 2 Tle 6-3: djustment Rtes on Cpitl... 2 Tle 6-4: Short-Run, Intermedite-Run, nd Long-Run Elsticities Tle 6-5: Distriution of the Firms which re Under-Utilizing Cpitl over Time.. 25 Tle 6-6: Short-nd Long-Run Scle nd Cost Elsticities, Period 987 to Tle 6-7: Short-Run nd Long-Run Scle nd Cost Elsticities, Period Tle 7-: Definition of the Inefficiency Components Tle 7-2: Definition of the Components of Dynmic Productivity Decomposition in the Presence of Inefficiency... 40

13 xiii Tle 7-3: Proportionl Growth of Output nd the Scle- nd Efficiency-Relted Components over Time, Tle 7-4: Components of Dynmic Totl Productivity Growth, (in Percentge) Tle 7-5: Components of Dynmic Totl Productivity Growth, (verge Vlues y Group of Firms in Percentge) Tle 7-6: Components of Dynmic Totl Productivity Growth, (verge Vlues y Group of High nd Low TFP Firms in Percentge) Tle 7-7: Components of Dynmic Totl Productivity Growth, (verge Vlues y Group of High nd Low Efficiency Firms in Percentge) Tle 7-8: Estimted Input Bis in Technologicl Chnge Tle -: Summry of Percentge Shre of Cpcity t U.S. Electric Utilities y Prime Movers, Tle -2: Lists of Electric Utilities in This Study Tle -3: verge nnul Output nd Vlue of Cpitl y Ech Utility from 986 to Tle -4: Summry Descriptive Sttistics of the Vriles Used for Estimtion in This Study y Ech Yer from 986 to Tle -5: ggregte Energy Produced of the Energy Producers over the Period of Study... 96

14 xiv Tle B-: Estimted Structurl Coefficients, GMM Estimtion (Full Model) Tle B-2: Estimtes of the Rtio of Optiml Cpitl to ctul Cpitl Reported y Firm for Tle B-3: Estimtes of the Short- nd Long- Run Mrginl Cost, verge Totl Cost, nd verge Vrile Cost Reported y Firm from Tle B-4: Estimtes of the Short- nd Long- Run Scle Elsticities nd Cost Elsticities Reported y Firm from Tle C-: Components of Dynmic Totl Productivity Growth y Firm for Tle C-2: Distriution of the Totl Fctor Productivity Growth... 25

15 xv CKNOWLEDGEMENTS From the time, when I decided to pursue my doctorl studies until the completion of this disserttion, I met mny people who inspired me oth professionlly, cdemiclly, nd spiritully. I would lie to express my deepest pprecition to my thesis dvisor, Prof. Spiro E. Stefnou, who is my role model for eing good scholr. Without his unwvering support, endless encourgement, invlule dvice, indeftigle ptience, nd wrmherted friendship, this disserttion could not hve een completed. Prof. Spiro E. Stefnou hs een much more to me thn Thesis dvisor nd Committee Chir. I m grteful to hve hd the opportunity to enefit from his guidnce. If I hve to list of wht I hve lerned during the period I discussed this disserttion with him, the list would e longer thn this disserttion. Thns for creting comfortle tmosphere for discussions nd llowing me to experiment nd lern y doing nd exploring things; Thns for discussing this disserttion with me with spirit, inspirtion nd joy; Thns for eeping my spirits nd ttitudes positive for my future creer; Thns so much for ming my lst yer of doing my PhD eing such memorle experience. I would lso lie to thn my committee memers, Dr. dm Z. Rose, Dr. Edwrd C. enice, nd Dr. Timothy. Considine for their vlule comments nd guidnce. I lso thn Dr. mes S. Shortle, Dr. Roderic Eggert, Dr. Zili Yng, nd Dr. Richrd L. Gordon, for providing stimulting clsses in res of energy, environmentl, nd minerl economics. I grtefully cnowledge the finncil supports from the Royl Thi Government, Deprtment of Sciences, Technology, nd Environment, nd from the Deprtment of Energy, Environmentl, nd Minerl Economics of the Pennsylvni Stte University in form of grdute ssistntship throughout my grdute study.

16 xvi dditionl thns re due to my grdute fellows in the Deprtment of Energy, Environmentl, nd Minerl Economics, especilly Tinwei (Tin) Zhng, for shring mny spects of life, love nd wor. The wrmherted friendship of Ttin (Tny) Borisov, grdute fellow from the Deprtment of griculturl, Environmentl nd Regionl Economics, will remin s prt of my memorle experience. I extend my thns to n Moyer nd Nncy Wrner for impeccle stff support nd for creting comfortle tmosphere for me throughout my doctorl studies. Specil thns re due to my Thi friends in Stte College for shring drems nd mition, especilly Hthirt (Bell) Mneetes for eeping me helthy nd eing such lovely sister; Dr. Nuntwn (Lynn) Silpngrmlers for continuous encourgement nd eing such gret listener. Thns for sending me n encourging messge in the right time. When the going gets tough, the tough gets going. I hve proved tht I m one of the tough ones. I dedicte this disserttion to my prents nd sisters. Without their endless love, continuous encourgement nd strong eliefs, I could not hve dremed of pursuing my journey wy from home. Thns for showing me how wonderful the world is. Thns for teching me how to give nd forgive, especilly eing le to fce impromptu situtions honestly nd grcefully. I m deeply indeted to ll of you for eeping me in good helth nd gret spirits. I lso thn my girlfriend, Pricht Mpun, for her immesurle ptience nd the strong elief tht love will me mircle thing. Lst ut not lest, I would lie to extend my profound pprecition to Professor Konrd G. Weil, n emeritus professor of Physicl Chemistry, Technicl University Drmstdt, Germny. His truly wrmherted friendship is the gretest nd most honorle present for me nd it will remin s prt of my unforgettle experience. Thns for lwys hving the door open nd llowing me to discuss nd shre mny spects of life. I m grdully soring his wonderful ttitude nd it will certinly influence me for the rest of my life. Our discussions out life, love, fmily, rt, sciences, economics, nd cooing will remin s prt of my memorle experience during my grdute study t the Pennsylvni Stte University.

17 Chpter Introduction Electricity deregultion nd restructuring re now on the policy gend in mny sttes. The sis for historicl regultion of the electricity industries hs een to del with nturl monopoly issues in the production of electricity. The first min step towrd deregultion ws the Pulic Utility Regultory Policies ct of 978 (PURP) pssed y the Congress which llowed independent genertors to sell their electricity to utilities t regulted rtes. In the cse of nturl monopoly industry, the regulted rtes re typiclly set equl to verge cost insted of mrginl cost of production to void incurring long-run losses. Under regultion, electric utilities will receive gurnteed profit for the genertion of electricity. Trditionlly, n electricity customer hs pid one regulted price for electricity to single verticlly integrted utility responsile for genertion, trnsmission, distriution, nd mreting. The 992 Energy Policy ct, followed y the Federl Energy Regultory Commission s (FERC s) Orders 888 nd 889 in pril of 996, expnded PURP s inititive y forcing utilities with trnsmission networs to deliver power to third prties t nondiscrimintory cost-sed rtes. These policy inititives recognize tht Smith (996) presents n interpretive overview of the history nd ptterns of regultion in the electric industry nd notes tht regultions in erly 900s were not motivted y consumerist response to monopoly pricing ut rther served to protect the industry from the competitive pricing which dominted t tht time.

18 2 while electricl trnsmission nd distriution remin nturl monopolies, competition in genertion is possile with new technology (e.g., gs turines) tht cn chieve optiml size t modest scle nd with open ccess to trnsporttion networs. Policies to open mrets led to new competitors in genertion nd mreting, with restructuring of the industry wy from the regulted, single-provider model. Deregultion in the electricity mrets hs to dte een incomplete with continued regultion in some of its segments. Under prtil regultion, electricity mrets re not relly deregulted ut restructured. Verticl integrtion hs diminished nd some stges of electricity provision must compete in the mret plce. Deregultion of energy genertion will provide importnt incentives for the efficient opertion of electricl genertors nd it should give firms the incentives to lower costs y improving technicl nd input lloctive efficiency to mximize their profits.. Sttement of Prolem Under the regultory er, electric utilities hd gurnteed profit for the genertion of electricity. This led to strong incentives to overinvest in cpitl s well s operting t n inefficient level of production which is of rod interest for reserchers nd policymers (verch nd ohnson (962), Crew nd Kleindorfer (2002), Grnderson nd Linvill (2002)). The level of inefficiency of electric utilities nd the forces driving inefficient levels of production electricity re criticl concerns. Since the underlying gol of the electricity deregultion pln is to improve the efficiency level of electric utilities, reserchers nd policymers focus on the

19 3 consequences of dopting this pln. The core issues ssocited with evluting the deregultion of electricity genertion surround mesuring the potentil sving in production costs due to the deregultion of this stge cross the country, the contriution of efficiency gins, nd evluting the djustment pth of n electric utility..2 Motivtion Understnding the cost structure of electric utilities in ddition to other economiclly meningful mesurements such s economies of scle, economies of scope, technology, technicl nd lloctive efficiencies, nd productivity growth in the electricity industry, provides insightful informtion for policy mers in deling with the issues relted to restructuring the electricity industry. The estimtion of the cost structure, scle economies, nd scope economies of electric power genertion in the United Sttes hs een extensively studied since the 960s. Nerlove (963) estimted scle economies in U.S. electric power genertion using Co-Dougls cost function nd found evidence of economies of scle in electric utility genertion in the 950s. Christensen nd Greene (976) used trnslog cost function to estimte economies of scle in U.S. electric power genertion nd found the supporting evidence of economies of scle in the genertion. These studies defined vrious functionl forms of cost functions nd pplied the conventionl estimtion pproch of the production nd cost structure. The shortcoming of this pproch is tht it ssumes tht ll producers operte efficiently.

20 4 The recent empiricl pplictions of electric power genertion relx this ssumption to mesure the inefficiency occurring in the production cost. Stochstic cost frontier nd shdow cost pproches re extensively pplied in the empiricl literture. The stochstic frontier pproch is cple of estimting the inefficiency level of ech producer. In contrst to the conventionl estimtion pproch of the production cost, the stochstic cost frontier pproch ssumes tht n efficiently operting producer is represented y the cost frontier. If producer s oserved costs re higher thn the frontier, tht devition is ttriuted, in prt, to inefficiency. Thus, devitions from the frontier re used to construct mesures of producer inefficiency. Schmidt nd Lovell (979) pplied this pproch on Co-Dougls functionl form. They estimted the cost inefficiency of the U.S. electric generting plnts nd decomposed the cost inefficiency into technicl nd lloctive inefficiency. However, their model specifiction could not e employed with other flexile functionl forms. The shortcoming of the stochstic frontier pproch is the computtionl difficulties for decompositions of economic efficiency in the estimtion. This shortcoming cn e remedied y using stochstic estimtion of the shdow price pproch originlly introduced y Hooper (965). The shdow price pproch estimtes technicl inefficiency modeled s fixed effect wheres lloctive inefficiency is mesured through input-specific prmeters tht scle mret prices. shdow cost function is expressed in terms of shdow input prices nd outputs, where shdow prices (internl to the firm) re defined s input prices forcing the techniclly efficient input vector the miniml cost solution for producing given output. Shdow prices my differ from

21 5 mret (ctul) prices. Using shdow cost system, where ctul costs nd input cost shres re expressed in terms of shdow costs, one cn chrcterize the cost structure of electric power genertion y estimting returns to scle, price elsticities of demnd for inputs, the lloctive nd technicl efficiencies, nd productivity growth. The shdow cost pproch hs importnt dvntges over the stochstic cost frontier pproch in providing econometric mesures nd decompositions of the economic efficiency. However, these two pproches to mesure the inefficiency occurring in the production cost re developed under the sttic context, where some inputs, referred to s qusi-fixed inputs, re constrined in the production decision ming model. Regultory, technologicl, mret, nd finncing constrints re fctors tht prevent the firm from reching new optiml levels of qusi-fixed inputs in the short-run. The shortcomings of the sttic pproches include ignoring the explicit the role of time, how the djustment of short-run qusi-fixed inputs to the oserved long-run level tes plce, nd how the opportunity cost of cpitl chnges fctor into the shdow cost of cpitl estimtes. The nlysis of the trnsition pth of qusi-fixed fctors towrd their desired long-run levels cn e remedied y explicitly incorporting costs of djustment for the qusi-fixed fctors. The underlying sic ide is tht the djustment process of qusi-fixed fctors genertes dditionl trnsitory costs nd the optiml intertemporl ehvior of the firm cn e solved y using the notion of djustment costs s mens to solve the firm s optimiztion prolem.

22 6 This study follows this notion of including djustment costs of qusi-fixed fctors into model of firm ehvior. The presence of djustment costs is sed on the underlying ssumption tht the stoc of qusi-fixed inputs is grdully ccumulted ecuse the economic environment plces higher cost on djusting the stoc rpidly thn slowly. The introduction of djustment costs leds to grdul djustment of qusi-fixed fctors only if the mrginl cost of djustment is incresing with the solute size of the djustment. The presence of djustment costs mes the process of improving firm s production dynmic one. The contriution of this study is to develop dynmic model of the cost minimizing firm nd to estimte the firms efficiencies. The sttic shdow cost pproch is generlized using the dynmic dulity model of intertemporl decision ming to estlish dynmic efficiency model of the cost minimizing firm. This study focuses on stem electric power genertion using fossil-fuel s primry fuel for mjor investor-owned utilities in the United Sttes. This genertion source is the dominnt prt of the electricity industry. Pnel dt on 72 electric utilities over the time period of will e the focus of the empiricl study. This pnel represents tht mture component of the industry which ccounts for the mjority of the power generted, trnsmitted nd mreted. The vilility of pnel dt generlly implies tht there re degrees of freedoms in the estimtion of prmeters nd such dt set permits the simultneous investigtion of oth the technicl chnge nd the technicl efficiency chnge over time. In ddition, the vilility of dt on outputs, input quntities, nd input prices enles seprtion of the cost inefficiency

23 7 into technicl nd input lloctive inefficiency components. The time period used in this study will cover the regultory er nd the restructuring er in the electricity industry which will enle us to understnd the impcts of electricity deregultion..3 Ojectives The specific ojectives of this study re summrized s follows: () to estlish dynmic model of cost minimizing firm nd to estimte the dynmic efficiency of U.S. electric utilities y use of the sttic shdow cost pproch nd the dynmic dulity model of intertemporl decision ming; (2) to estimte nd decompose the cost inefficiency into four components: lloctive nd technicl inefficiencies of net investment demnd nd vrile inputs demnd; (3) to chrcterize the dynmic structure of electric power genertion using input prices elsticity, scle nd cost elsticities, nd cpcity utiliztion under dynmic djustment; (4) to develop theoreticlly consistent method in mesuring the decomposition of the dynmic totl fctor productivity growth in the presence of inefficiency; (5) to exmine how the electric utilities will prticipte in the deregultion of the production of electricity; in prticulr, to evlute how different

24 8 electric utilities will perform tht re locted within or outside of sttes with the restructuring pln..4 Chpter Orgniztion The rest of this thesis is orgnized s follows. In Chpter 2, literture reviews of development of dynmic fctor demnds nd mesurements of productivity, efficiency nd totl fctor productivity growth re presented. Chpter 3 contins theoreticl cground on the economics of efficiency nd dynmic fctor demnds. The integrted theoreticl nd econometric models of dynmic efficiency model re developed in Chpter 4. In Chpter 5, cground on the U.S. electricity industry nd issues surrounding in restructuring the U.S. electricity industry re provided. The primry sources of dt used to construct the dt set nd ey ssumption underlying the construction re lso discussed in Chpter 5. Chpter 6 contins the estimtion results of the dynmic efficiency model followed y the estimtion results of dynmic totl fctor productivity growth in the presence of inefficiency in Chpter 7. Finlly, Chpter 8 summrizes the empiricl results nd offers suggestions for future reserch.

25 9 Chpter 2 Literture Reviews This chpter is orgnized into four sections. Section 2. reviews the development of dynmic fctor demnds. Section 2.2 provides n overview of the mesurement of productivity nd efficiency. Section 2.3 reviews the development of mesurement of totl fctor productivity growth decomposition. Finlly, Section 2.4 surveys empiricl studies of the mesurement of productivity nd efficiency in the production of electricity industry. 2. Development of Dynmic Fctor Demnds The erliest dynmic energy demnd model ws developed y using the Koyc prtil djustment with single eqution of energy demnd. The model relted the ctul chnge in the quntity of energy demnd etween the current time periods nd the previous time periods is equl to proportion of the long-run chnge. This prtil djustment mechnism is esily implemented empiriclly. However, it is not sed on economic optimiztion nd fils to ccount for interrelted spects of the djustment process, prticulrly the extent of disequilirium in the cpitl fctor demnd eqution. orgenson (963) formulted the flexile ccelertor model which provides systemtic elimintion of the discrepncy etween the trget nd the ctul level of

26 0 qusi-fixed input through the prtil djustment process. Ndiri nd Rosen (969) susequently introduced the interrelted disequilirium in one fctor mret to the extent of disequilirium in other fctor mrets. In ddition, they extended nd implemented empiriclly the flexile ccelertor model of orgenson (963) to the cse of multiple qusi-fixed fctors. lthough their model involved systems of interrelted disequilirium equtions rther thn single demnd eqution, functionl form of the Co-Dougls technology ws ssumed nd prtil djustment coefficients were constrined to e constnt prmeters into the model. These restrictions often yielded implusile empiricl results. nother lterntive pproch of dynmic fctor demnd ws developed y Lu (976) nd McFdden (978). Their pproch ws sed on the notion of restricted vrile cost, or restricted vrile profit function. For instnce, the restricted vrile cost function reflects production or technologicl constrints fcing the firm when output nd certin input quntities clled qusi-fixed inputs re fixed in the short-run. This frmewor of the firm s optimiztion prolem is referred to s temporl equilirium or prtil sttic equilirium. This pproch hs some dvntges. It genertes short-run demnd equtions for vrile input nd long-run demnd equtions for oth vrile nd qusi-fixed inputs. Further, it permits clcultion of n economiclly meningful mesure of cpcity utiliztion. However, the temporry equilirium frmewor crries some shortcomings ecuse it does not explicitly recognize the role of time. In ddition, the djustment of short-run qusi-fixed inputs

27 to the oserved long-run level is not explicitly modeled nd it does not indicte why certin inputs re fixed in the production process. The sence of n explicit nlysis of the trnsition pth of qusi-fixed fctors towrd their desired long-run levels cn e remedied y explicitly incorporting costs of djustment for the qusi-fixed fctors. The underlying ide is tht the djustment process of qusi-fixed fctors genertes dditionl trnsition costs nd the optiml intertemporl ehvior of the firm cn e solved y using the notion of djustment costs s mens to solve the firm s optimiztion prolem. The frmewor of this firm s optimiztion prolem is referred to s intertemporl equilirium. 2.. Intertemporl Firm Behvior The frmewor of the optiml intertemporl ehvior of the firm using the notion of djustment costs s mens to solve the firm s optimiztion prolem ws first introduced y Eisner nd Strotz (963) nd susequently developed y Lucs (967), Gould (968), nd Tredwy (969, 970, 97, 974). The presence of djustment costs formlizes the process of chrcterizing firm s dynmic production decisions. In the presence of costs of djustment for the qusi-fixed fctors, firm fces dditionl costs of the djustment costs of qusi-fixed inputs eyond cquisition costs in the decision ming process. There re two sources of djustment costs discussed in the literture. Internl djustment costs re the result of reduction in productivity which occurs when more of qusi-fixed fctor is sored or relesed too quicly.

28 2 Costs include temporry loss in output production. Internl djustment costs suggest specifiction of the firm s production technology y including the vector of rtes of chnge of the qusi-fixed inputs into the production function. Externl djustment costs rise from mret forces or contrctul oligtions. Exmples of externl djustment costs re expnsion plnning fees nd imperfect cpitl goods mrets. The presence of djustment costs is lid on the underlying ssumption tht the stoc of qusi-fixed inputs is grdully ccumulted ecuse the economic environment plces higher cost on djusting the stoc rpidly thn slowly. The introduction of djustment costs into model of firm ehvior leds to grdul djustment of qusi-fixed fctors only if the mrginl cost of djustment is incresing with the solute size of the djustment. The dynmic economic prolem fcing the firm cn e ddressed y chrcterize firm investment ehvior s the firm seeing to minimize the present vlue of production costs over time horizon. Production costs rise from purchsing new inputs, oth vrile nd qusi-fixed inputs. Units of the qusi-fixed inputs re cquired oth for enlrging of the existing productive cpcity nd for replcing of worn-out units. The dynmic model for the study of optiml input decisions hs een employed for oth priml nd dul representtions. The priml pproch develops the consistent system of input demnd equtions y directly deling with the priml functions such s the production function, the restricted cost function conditionl on

29 3 investment, nd the cost of djustment function, wheres the dul pproch dels with the indirect or vlue function of the dynmic progrmming eqution. The priml pproch ws developed y Tredwy (97, 974) nd hs een pplied to oth U.S. ggregte mnufcturing nd to two-digit industry dt y Berndt, Fuss, nd Wvermn (979). Numerous empiricl pplictions of the priml pproch include Morrison (983, 986), Wtins nd Berndt (992), nd Lee nd Kwon (994). Morrison (983) nd Wtins nd Berndt (992) conducted joint tests of specific djustment cost structures, optiml trjectories of the flexile ccelertor type, nd of the sttic expecttions hypothesis using qudrtic cost function. Morrison (983) performed the tests on U.S. ggregte mnufcturing dt wheres Wtins nd Berndt (992) tested their models on Cndin dt for ggregte mnufcturing. Morrison (986) extended the study of Morrison (983) y relxing the sttic expecttion ssumption. Lee nd Kwon (994) estimted the complete dynmic fctor demnd system under lterntive schemes of expecttions formtion presented in Morrison (986) to compute cpcity utiliztion indexes for the Koren mnufcturing sector. shortcoming of priml pproch is tht it is limited to modeling only one qusi-fixed input, or ssuming independent djustment etween two or more qusi-fixed input. This shortcoming cn e remedied y using the dul pproch. The dul pproch llows modeling more thn one qusi-fixed input nd to test for independent djustment rther thn ssuming it holds.

30 4 The theory of intertemporl dulity ws improved upon y McLren nd Cooper (980) nd Epstein (98). This theory represents n lterntive nd powerful method to solve intertemporl optimiztion prolems y using the optiml vlue function of the dynmic progrmming eqution pproch. The first empiricl ppliction sed on this theoreticl pproch ws study on U.S. mnufcturing y Epstein nd Denny (983) which ssumed cpitl nd lor to e qusi-fixed inputs nd estimted flexile ccelertor system derived from qudrtic optiml vlue function. Tests of symmetry of the Hessin of the vlue function, symmetry of responses in short-run nd long-run fctor demnds, vriility of ech qusi-fixed input considered, nd independent dynmic djustments re evluted. Mny of the empiricl pplictions of intertemporl dulity theory re pplied in the griculturl economics literture. They re different ccording to how the flexile functionl forms of the vlue function of the dynmic progrmming eqution re defined nd wht industries re studied. Choices of the flexile functionl forms pplied in the empiricl literture include qudrtic, trnslog, or Generlized Leontief (GL). The flexiility stnds for the ility of imposing s few priori restrictions s possile upon the nlyticl representtion of the production structure. The qudrtic, trnslog, nd GL functionl forms cn ccommodte multiple outputs without necessrily violting curvture conditions nd cn mintin the symmetry property. The trnslog nd GL functionl forms permit imposition of the liner homogeneity property nd to test for the liner homogeneity hypothesis where the qudrtic functionl form does not. The qudrtic functionl form hs some dvntges over other functionl forms. It is

31 5 consistent with ggregtion cross production technology (Epstein nd Denny, 983) nd it is self-dul in the sense of Lu (976) llowing the possiility of solving the vrile cost function nlyticlly for the ssocited qudrtic production function nd vice vers. On the other hnd, the trnslog functionl form cnnot yield nlyticl or closed-form expressions for the levels of qusi-fixed inputs. The GL functionl form llows closed-form solution for the equilirium level of qusi-fixed input (Morrison, 988) nd it is good pproximtion of technology when limited input sustitution possiilities existed (Cve nd Christensen, 980). Exmples of the econometric estimtion of dynmic dul models include Vsvd nd Chmers (986), Howrd nd Shumwy (988), Luh nd Stefnou (99, 993), Fernndez-Cornejo et l. (992), nd Mner (994). Vsvd nd Chmers (986) used qudrtic vlue function nd conducted severl tests such s symmetry in short-run nd long-run fctor demnd nd independent dynmic djustments. They performed the tests on U.S. griculturl dt. Howrd nd Shumwy (988) conducted similr tests nd used the modified Generlized Leontief vlue function on U.S. diry industry dt. Luh nd Stefnou (99) lso used the modified Generlized Leontief vlue function to mesure totl fctor productivity growth in U.S. griculture dt. Fernndez-Cornejo et l. (992) used normlized qudrtic vlue function on Germn griculture dt to estimte mesures of scle nd scope economies. Mner (994) used qudrtic optiml vlue function to investigte the production structure of the Itlin mnufcturing sector.

32 6 2.2 Mesurement of Productivity nd Efficiency In the theory of production, productivity nd efficiency re used to mesure the performnce of firms which convert inputs into outputs. Productivity is defined s the rtio of the outputs produced to the inputs used y firm in the production process. production frontier is defined s the mximum output ttinle from ech input level which represents est-prctice technology of firms in the industry. Firms in tht industry who re reltively technicl efficient will operte on the frontier wheres firms who re reltively technicl inefficient will operte eneth the frontier. The first nd foremost of productivity mesurement ws the use of the Fisher (922) nd Tornqvist (936) indices. It ws susequently developed nd sed upon the ide of Mlmquist (953) nd Shephrd (953) who independently introduced the notion of distnce function. Efficiency mesurement egn with Frrell (957) who presented computtionl mesures for productive inefficiency tht drew upon the wor of Dereu (95) nd Koopmns (95). Frrell (957) chrcterized totl economic efficiency of firm consisting of two components: technicl efficiency nd lloctive efficiency. Technicl efficiency reflects to the ility of firm to otin mximl output from given input vector ( output-oriented ) or the ility of firm to minimize input use in the production of given output vector ( input-oriented ). The nlysis of technicl efficiency cn te on output-oriented or input-oriented focus of technicl efficiency. lloctive efficiency reflects the ility of firm to use the inputs in optiml proportions given their respective prices nd the production

33 7 technology. Frrell suggested the use of either non-prmetric piece-wise-liner convex isoqunt or prmetric function with the Co-Dougls form to estimte the production frontier nd to mesure the firm s inefficiency level s the derivtion from the frontier. There re vrious methods to mesure the performnce of firms. Ech method differs ccording to type of techniques used, type of dt ville, nd the ssumptions mde regrding the economic ehvior of decision mers nd the structure of the production technology. The mesurement of productivity nd efficiency cn e clssified into the non-prmetric nd the prmetric pproches. These two pproches differ in vrious wys. The non-prmetric pproch does not require specific functionl form representing the underlying production process to e imposed on the dt while the prmetric pproch requires the specifiction of functionl form. The non-prmetric pproch typiclly does not te sttisticl noise into ccount, which consequently provides inccurte efficiency mesures, while the prmetric pproch esily ccommodtes sttisticl noise Non-prmetric pproch to the Mesurement of Productivity nd Efficiency The non-prmetric pproch widely pplied in the literture is clssified into two methods. One is totl fctor productivity (TFP) index nd the other is dt

34 8 envelopment nlysis (DE). The TFP index method ssumes ll firms re techniclly efficient. It is often pplied to time-series dt or pnel dt nd provides mesures of TFP. On the other hnd, the DE method does not ssume ll firms re techniclly efficient. It is most often pplied to cross-sectionl dt nd provides mesures of reltive efficiency nd scle efficiency mong those firms. If pnel dt of costs, input quntities, nd input prices re ville, the DE cn e used to mesure oth technicl nd efficiency chnge nd to estimte lloctive efficiency nd congestion efficiency. The TFP index is defined s the rtio of n ggregte output quntity index to n ggregte input quntity index. Productivity growth occurs when n index of outputs chnges t different rte thn n index of inputs. The input nd output quntity index numers, nd productivity indices re ll sed on the ides of Mlmquist nd the distnce function pproch outlined in Mlmquist (953). theoreticl frmewor for the mesurement of productivity growth using index numer pproch is drwn upon erlier studies y Cves, Christensen nd Diewert (982), Diewert (992), Färe, Grossopf nd Roos (997), nd Fǿrsund (997). There re two theoreticl pproches to mesuring productivity growth. The first pproch is Mlmquist Productivity Indices nd the second pproch is Hics-Moorstien Productivity Indices. Cves, Christensen nd Diewert (CCD) (982) developed Mlmquist Productivity Index Numer pproch using the distnce function introduced y Mlmquist (953) to mesure the chnge in productivity. Under the CCD pproch, there re two pproches used to mesuring

35 9 productivity growth. One is referred to s the output-oriented mesure which focuses on the mximum level of outputs tht could e produced using given input vector nd given production technology reltive to the oserved level of outputs. The other is referred to s the input-oriented mesure y exmining the fesile reduction in input use which is fesile given the need to produce given level of output under reference technology. Diewert (992) presented the Hics-Moorstien pproch which drws upon the erlier wors y Hics (96) nd Moorstien (96). Hics-Moorstien Productivity Indices re defined s the rtio of output index numers to input index numer. The respective indices re ll defined using Fisher, Tornqvist or other indices. Bjure (996) defined the respective indices using the Mlmquist pproch nd referred productivity index s the Mlmquist Totl Fctor Productivity Index. The DE pproch constructs the frontier over the dt nd mesures efficiency reltive to the constructed frontier using liner progrmming techniques. The DE pproch ws introduced y Chrnes, Cooper nd Rhodes (978) who drew upon erlier efforts y Frrell (957), Boles (966), nd frit (972). Chrnes, Cooper nd Rhodes (978) proposed n input-oriented constnt returns to scle model. Lter, Bner, Chrnes, nd Cooper (984) extended the constnt returns to scle DE model of Chrnes, Cooper nd Rhodes to ccount for vrile returns to scle. The vrile returns to scle ssumption is pproprite when firms fce imperfect competition or constrints in the production nd my not operte t n optiml scle. The vrile returns to scle DE model llows the clcultion of the scle

36 20 efficiencies effect. The DE pproch cn e ctegorized ccording to the type of vriles ville (input nd output quntities nd prices), nd ccording to the type of dt ville (cross-sectionl, or pnel dt). With only quntities ville, technicl efficiency cn e estimted, wheres lloctive efficiency cn e mesured if oth quntities nd prices re ville nd ehviorl ojective, such s cost minimiztion nd revenue mximiztion is mintined. Cost minimiztion nd revenue mximiztion together imply profit mximiztion. Profit efficiency using DE methods hs een rrely studied in empiricl nlysis with recent study presented y Färe, Grosshopf nd Weer (997). Comprehensive reviews of the DE model re presented y Seiford nd Thrll (990), li nd Seiford (993), Lovell (993, 994), Chrnes et l (995), nd Seiford (996) Prmetric pproch to the Mesurement of Productivity nd Efficiency The prmetric technique extensively pplied in the literture is clssified into two methods. First is the conventionl lest squre econometric model nd second is stochstic frontier model. The conventionl lest squre econometric model method ssumes ll firms re techniclly efficient. It cn e pplied to cross-sectionl dt, time-series dt, or pnel dt nd provides mesures of technicl chnge nd scle economies. On the other hnd, the stochstic frontier model method does not ssume tht ll firms re techniclly efficient. It is most often pplied to cross-sectionl dt nd provides mesures of reltive efficiency nd scle economies mong those firms. If pnel dt of costs, input quntities, nd input prices re ville, the stochstic

37 2 frontier cn e used to mesure oth technicl chnge nd efficiency chnge nd to decompose efficiency into technicl efficiency nd lloctive efficiency. The conventionl lest squre econometric nlysis hs een extensively used for oth priml nd dul representtions of production technologies. The production function is used for the priml representtion wheres cost nd profit re used s the dul counterprt. Economic estimtion of production, cost or profit function requires the selection of n pproprite functionl form. These two representtions differ ccording to the type of dt ville nd vrious ehviorl ssumptions. The priml pproch only requires dt on quntities of inputs nd outputs while the dul pproch lso requires price dt. nother gret dvntge of the dul pproch is the ility to del with multi-output technologies. The stochstic frontier nlysis constructs the frontier function over the dt nd estimtes efficiency reltive to the constructed frontier using econometric techniques. Stochstic frontier nlysis hs some dvntges reltive to the DE nlysis for two resons. First, it does te sttisticl noise into ccount, which consequently provides ccurte efficiency mesures. Second, it llows performing tests of hypotheses regrding the existence of inefficiency nd the structure of the production technology. Stochstic frontier function nlysis ws independently proposed y igner, Lovell nd Schmidt (977), Bttese nd Corr (977), nd Meeusen nd vn den Broec (977) who drew upon erlier wors y Frrell (957), nd igner nd Chu (968). igner nd Chu (968) proposed deterministic specifiction of production

38 22 frontier. The deterministic frontier specifiction of igner nd Chu (968) involved only non-negtive rndom vrile ssocited with technicl inefficiency in production of firm. Lter, igner, Lovell nd Schmidt (977) nd Meeusen nd vn den Broec (977) simultneously introduced stochstic specifiction of production frontier. Their stochstic frontier specifictions involved stochstic error term ssocited to rndom shocs outside the control of producers tht cn ffect output nd non-negtive rndom vrile ssocited with technicl inefficiency in production of firm. The stochstic frontier specifiction hs een extensively used compred to the deterministic frontier specifiction ecuse the former does te sttisticl noise into ccount resulting in more ccurte specifiction nd consequently more ccurte efficiency estimtes. The stochstic frontier nlysis hs een employed for oth priml nd dul representtions of production technologies. The priml representtion uses production nd distnce functions, while cost nd profit functions re used s the dul counterprt. These two representtions differ ccording to the ehviorl ssumption mde, the type of informtion otined, nd the type of dt ville. The priml pproch does not require the imposition of ehviorl ojective on producers, wheres the dul pproch is sed on ehviorl ssumptions, such s cost-minimiztion, or profit-mximiztion. The priml pproch llows to estimte technicl efficiencies, wheres the dul pproch llows to estimte cost efficiencies nd to decompose cost efficiencies into technicl efficiencies nd lloctive efficiencies. The priml pproch requires dt only on quntities, wheres the dul pproch requires dt on oth

39 23 quntities nd prices. stochstic production frontier of the priml pproch does not llow distinguishing etween vrile inputs nd qusi-fixed inputs ecuse it trets ll input eqully. On the other hnd, stochstic cost frontier of the dul pproch llows treting vrile nd qusi-fixed inputs differently. The dul pproch of stochstic frontier ws introduced y Schmidt nd Lovell (979), lthough deterministic cost frontier ws previously estimted two yers erlier y Fǿrsund nd nsen (977). Direct estimtion of the dul profit frontier hs een studied lter y Holls nd Stnsell (988), li nd Flinn (989), tinson nd Kervliet (989), Kumhr, Ghosh nd McGucin (99), Kumhr nd Bhttchryy (992), nd Berger nd Mester (997). Erlier studies of stochstic frontier nlysis were ddressed using with cross-sectionl dt on N producers. The cross-sectionl stochstic production frontier function of igner, Lovell, nd Schmidt (977) ssumed non-negtive rndom vrile ssocited with technicl inefficiency following the hlf-norml nd the exponentil distriutions. The specifiction of these distriutionl ssumptions ecme one of the min criticisms in the stochstic frontier nlysis tht there is no priori justifiction for the selection of ny prticulr distriutionl form for the technicl inefficiency term. The criticism led to the development of the specifictions of more generl distriution forms presented in the following studies. Stevenson (980) introduced the truncted-norml model, nd Greene (990) proposed the two-prmeter gmm model. Stevenson pplied the truncted-norml model to fit oth production nd cost functions into the empiricl nlysis, wheres Greene