光伏补贴政策设计与评估--以美国加州为例

Contents

PrefaceⅠ

List of TablesⅨ

List of FiguresⅪ

Chapter 1 Introduction1

Chapter 2 Policy Introduction: theCalifornia Solar Initiative7

1)The Joint Staff Report8

2)Megawatt-Triggering Mechanism10

3)Incentive Application Process13

Chapter 3 Optimal Subsidy Design withStochastic Learning: A Dynamic

Programming Evaluation of the CaliforniaSolar Initiative17

1)Introduction17

2)The California Solar Initiative: Policyin Retrospect21

(1)CSI Target and Budget Setting21

(2)Megawatt-Triggering Mechanism22

(3)CSI Performance23

3)Modeling and Parameterization25

(1)Model Setup25

(2)Parameterization28

4)Results39

(1)Analytic Results39

(2)Deterministic Case41

(3)Stochastic Case51

5)Conclusions57

Chapter 4 Incentive Pass-through forResidential Solar Systems in California60

1)Introduction60

2)Literature Review63

3)Methods and Data67

(1)Structural Modeling68

(2)Reduced-form Regression73

(3)Data74

4)Results81

(1)Structural Modeling81

(2)Reduced-form Approach87

5)Conclusions91

Chapter 5 Analyzing Incentive Pass-throughfor the California Solar Initiative:

A Regression Discontinuity Design95

1)Introduction95

2)CSI Policy Design and Suitability for RDAnalysis98

3)Methods and Data100

(1)Methods100

(2)Data104

4)Results112

(1)Time Discontinuity112

(2)Geographic Discontinuity122

5)Conclusions127

Chapter 6 Conclusion130

Appendix134

Bibliography137

Chapter 1 Introduction

 Human-induced climate change, with itspotentially catastrophic impacts on weather patterns, water resources,ecosystems, and agricultural production(IPCC WG2, 2014), is the toughest globalproblem of modern times(Dow and Downing, 2011). Based on the latest projectionby the Intergovernmental Panel on Climate Change(IPCC), global surfacetemperature has been increasing almost linearly in the past four decades or so,and the temperature change is highly likely to exceed 2℃ by the endof the 21st century(IPCC WG1, 2014). Such change is likely to cause significantglobal GDP losses; the well-known Stern Report(Stern, 2007) estimated that fora global mean temperature change of 2℃~3℃, thepotential global GDP loss would be around 1%~2%. This number may look small,however, no single country wishes to bear the burden alone.

While the concentration of greenhousegases(GHGs) from the human activities is the largest driver of the observedclimate change(EPA, 2014a), tracing the sources of GHGs reveals that in the UnitedStates, electricity generation produces the largest share of GHGs. In 2012,this sector emits 32% of GHGs in the United States, followed by thetransportation sector at 28%(EPA, 2014b). Obviously, different electricitygeneration technologies tend to have very different emission rates. Based on arecent review report(WNA, 2011) that summarizes life-cycle assessments of theGHG emission intensity for different generation technologies, solarphotovoltaic(PV) only emits 85 tons of CO2e per GWh, while the numbers fornatural gas and coal are 500 and 888 respectively.

Unfortunately, electricity generation costsdo not necessarily reflect the differences among technologies in terms of GHGemission intensity, within a market where there is no price for carbon or GHGs.This is the so-called environmental externality problem, i.e. the GHG emittersdo not need to pay for the environmental damages that they cause. A naturalcure for this externality problem is to put a price on carbon or all GHGs, i.e.a Pigovian tax(Pigou, 1920). Nevertheless, few countries have chosen this path;instead, most of them have come to support renewable energy technologiesdirectly. As pointed out in the IPCC mitigation report, impeding catastrophicclimate change necessitates the widespread deployment of renewable energytechnologies for reducing the emissions of heat-trapping gases, especiallycarbon di-oxide(CO2)(IPCC WG3, 2014).

The solar PV industry has been growing veryrapidly in the last decade. According to the International Energy Agency(IEA),since 2000 solar PV has had the fastest growth rate among renewable energytechnologies worldwide(IEA, 2010). While the global annual installed PVcapacity was less than 0.3 gigawatts(GW) in 2000, this number surpassed 38GW in2013(EPIA, 2014). Reflecting the rapid growth in deployment, global investmentin solar energy technologies has been over $100 billion since 2000(Statista,2014). Deployment in the United States has also grown rapidly from around0.004GW of newly installed capacity in 2000 to more than 4GW installed in 2013alone(Sherwood, 2013; SEIA/GTM, 2014).

A key driving force behind the growth ofsolar PV has been the myriad of government incentive programs promoting solardeployment(Arvizu et al., 2011; Kirkegaard et al., 2010; REN21, 2014; Timilsinaet al., 2011), often motivated by a desire to address various market failuressuch as: environmental externalities as mentioned above(Baumol and Oates, 1988;Bezdek, 1993; Painuly, 2002; Stavins, 2008), learning-by-doing, innovationspillover effects, and peer effects in the PV industry(Arrow, 1962; Gillinghamand Sweeney, 2012; McDonald and Schrattenholzer, 2001; van Benthem et al.,2008; Verdolini and Galeotti, 2011). Other factors driving policy decisions tosupport solar include the potential benefits of energy resource diversity(i.e.energy security) and the potential of new jobs and increased economic activityin the solar sector(Fischer and Preonas, 2010). In addition, since most of theincentive programs affect the demand side, the induced innovation engendered bythese demand-pull policies brings in additional benefits to society(Hickes,1932; Jaffe and Newell, 2002; Lanzi and Sue Wing, 2011; Nemet, 2009a; Popp etal., 2010). As solar deployment increases rapidly due to these demand-pullpolicies, solar modules, the key component of a PV system, have experienced acost reduction by a factor of over 100 since the 1950s(Maycock, 2002; Nemet,2006), with recent prices as low as 60~70 cents per Watt(GTM Research, 2014).

Direct policy instruments that supportsolar PV deployment can take many forms, including feed-intariffs(FiT),renewable portfolio standards, investment tax credits(ITC), upfront rebates,net metering, favorable financing, mandatory access, and public investment. Indirectpolicy tools also exist such as carbon tax and cap-and-trade. The relativemerits of these instruments have been broadly studied and debated(Fischer andNewell, 2008; Fullerton and Melcalf, 2001; Nordhaus, 1992; Pizer, 1999;Vollebergh and van der Werf, 2014; Weitzman, 1974). While recognizing thecomplexity of the problem, this dissertation decides to focus on one of thepolicy tools–the upfront rebates program, though from several differentperspectives. The understanding of the design features and effectiveness ofthis policy tool provides a strong foundation to study inter-policyrelationships in the future.

Upfront rebates directly speak to the highcapital cost problem facing potential PV adopters, which is one of the majorbarriers to the diffusion of renewable energy technologies(Beck and Martinot,2004; Hoff, 2006; Sawin, 2004; Verbruggen et al., 2010). Though the average PVinstallation price has come down dramatically in recent years(Barbose et al.,2014), a typical residential PV system in the U.S.

(4kW) still costs around $20,000 on apre-rebate basis. In the U.S., the upfront rebate only exists at the statelevel or below, and governments usually base their rebate on PV systemproduction(i.e. performance), capacity, or both. Production-based subsidiesencourage better siting, configuration, and operation andmaintenance(O&M), thus maximizing potential production by tying theincentives to system performance; whereas capacity-based subsidies address thecapital cost problem directly and play a significant role in attracting lowerPV capacity customers and small projects(Barbose et al., 2006; Black, 2006;Connor et al., 2009; Hoff, 2006; IPCC, 2011).

For both production-based andcapacity-based incentives, setting an appropriate incentive level is always amajor challenge for policymakers. Since the PV technology is evolving rapidly,it becomes difficult to set up the incentive at the right level: too high anincentive level would attract too many applications leading to a run on the program’sbudget, while too low a level would do little to induce market growth. Chapter3 tackles this problem in the framework of dynamic programming, which has beenapplied before in the literature to tackle similar problems. I use the biggeststate-level rebate program in the Unites States, the California SolarInitiative(CSI), as the central example for its empirical focus. Theavailability of rich data for CSI and its significant scale offer a goodopportunity to examine the problem in detail. While Chapter 2 introduces theCSI policy, Chapter 3 provides several key insights regarding subsidy policydesign focusing on its cost effectiveness.

Another important perspective on thequestion of subsidy policy design looks at the redistribution effect. This effectis concerned with where the subsidy finally ends up, and whether it benefitsconsumers or suppliers more. This is an important and much studied question inpublic economics, i.e. the so-called subsidy incidence or incentivepass-through question. However, despite CSI’s significant program budget(over$2 billion), there are few studies that carefully study at the incentivepass-through question for CSI. Chapter 4 fills in the gap and adopts twoapproaches to answer this question: structural modeling based on the conductparameter approach and a reduced-form regression analysis. In my analysis Iview these two approaches as being complementary to each other, since differentunderlying assumptions and data requirements are involved.

In Chapter 5 I employ an as-if naturalexperiment design to re-examine the incentive pass-through question using aregression discontinuity(RD) design. Under certain assumptions, the RD designcould improve the internal validity of research similar to randomized controlexperiments(Imbens and Lemieux 2008; Lee, 2008). That is one of major reasonsfor the increasing popularity and adoption of this method in economics andother areas in the social sciences. As for CSI, the pre-determined incentivelevel stepwise changes and the geographic borders between two neighboringutilities provide good opportunities to apply the RD design. As a result, thederived incentive pass-through rate can be claimed as causal effects, a furtherrobustness check to estimates from Chapter 4, while the latter complementsChapter 5 by providing external validity to the pass-through results. Theresults from these two chapters have direct implications for subsidy policydesign. A complete pass-through rate indicates that the subsidy has benefitedfully the intended recipient, i.e. the consumers, and that the induced marketcompetition by the subsidy policy is probably high.

Chapter 6 concludes the dissertation, whilesynthesizing findings from the core Chapters 3~5, upon which the dissertationis centered. It further discusses fruitful research directions as next steps.The conclusion is kept short by choice, since there are correspondingconclusion sections in each of the core chapters. Overall, this dissertationmakes both empirical and methodological contributions to the public policyliterature, especially in policy design and evaluation. First of all, it servesas a thorough empirical study of incentive policy design, from both acost-effectiveness perspective(Chapter 3) and a redistribution point of view(Chapter4 and 5). Second, methodologically Chapter 3 extends the deterministic dynamicprogramming framework to further incorporate the stochastic learning-by-doingphenomenon. The considerations of various PV demand functional forms and policyflexibility as well as policy certainty are also new to the literature. Third,Chapters 4 and 5 examine the incentive pass-through question from multipleangles, and they are quite comprehensive in looking at this specific problemfor solar PV. Lastly, Chapter 5

also develops several adaptions of the RDdesign to fit the PV price data, as they proved to be important in removingpotential biases in the estimation process.