1. The Learning Process and Technological Change through International Collaboration: Evidence From China's CDM Wind Projects Tian Tang David Popp Maxwell School Syracuse University Public Policy in Asia, Singapore, May 26-27, 2014

2. Research Question Research Question: How does the learning process lead to technological change in wind power? Technological Change: -Overall production cost: reduction in unit cost of wind power -Wind farm installation: reduction in unit capital costs -Wind farm operation: capacity factor Learning Process: How the knowledge related to wind power is acquired and diffused among project participants Case: China’s wind power projects supported by the Clean Developed Mechanisms (CDM)

3. Background: CDM and China’s Wind Industry CDM: A project-based carbon transaction mechanism under the Kyoto Protocol that allows developed countries with emission constraints to purchase emission credits by financing projects that reduce carbon emissions in developing countries. Goals of CDM: -Help developing countries reduce carbon emissions -Stimulate sustainable development in developing countries through technology transfer from developed countries The Role of CDM in China’s Wind Industry -China has actively engaged in CDM since 2002 and used it to provide financial support for over 80% of wind projects - Standardized and transparent process and validated data

4. Background: International Collaboration in CDM Wind Projects Engage both the public and private sectors of the carbon trading countries, and international organizations CDM Project hosting country Central and local government, wind power companies, carbon trade consultants Emission credits buying country Central government, investment banks or carbon trading firms International organizations administering and monitoring CDM process Executive Board of CDM, 3 rd party validating and monitoring agencies

5. Theory: Learning Process and Technological Change Following the technological learning and collaboration theories, we identify the following channels of learning that could lead to the reduction of electricity production cost: Project Developer - Site selection - Construction - Operation - Connecting with the grid for delivery Project Developer - Site selection - Construction - Operation - Connecting with the grid for delivery Wind Turbine Manufacturer -Research and development -Produce wind turbines Wind Turbine Manufacturer -Research and development -Produce wind turbines Installation, Training, O&M Learning through R&D (Learning by Searching, LBS) Learning by doing (LBD) Learning by interacting (LBI) Learning by doing Knowledge spillovers

6. Data and Empirical Model Unit of Analysis: CDM wind power project Data -Pooled cross-sectional -510 registered CDM wind projects in China that started from 2002 to 2009 -Including 92 developers and 33 turbine manufacturers Sources: 1)Validated CDM project design document and its attached financial analysis spreadsheet for each project 2)Yearbook from Chinese Wind Energy Association 3)DELPHION patent database

7. Data and Empirical Model Dependent Variables: 1) Projected unit cost of electricity production of project i started construction in year t -Levelized cost 2) Projected unit capital cost of project i started in year t 3) Projected capacity factor of project i started in year t

8. Data and Empirical Model Explanatory Variables: Learning Effects LBS LBD Spill- over LBI

9. Data and Empirical Model Full Model: Manufacturer’s knowledge stock Manufacturer’s experience and developer’s experience Spillover from the industry Interacting experience Wind resource

10. Descriptive Statistics: Projected Unit Cost of Electricity Production Decreases from 2005- 2009 Unit cost: 12.1% Unit capital cost: 10.8% Unit O&M cost: 17.4%

11. Empirical Results Effect of aggregate level experience Effect of developer’s and manufacturer’s internal experience v. spillover effects Effects of interacting experience and other channels of learning

12. Empirical Results 1.Learning by searching -Wind projects benefit from the knowledge stock of their turbine manufacturers. -The effect is small though.

13. Unit Production Costs Note: 1. Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1. 2. Industrial level experience omitted in model (2), (6) and (7) because the sum of different levels of experience equals to the total industry-wide capacity in a given year, which is correlated with the year dummies. Dependent Variable:(1)(2)(3)(4)(5)(6)(7) ln (unit_cost)Aggregate level Internal experience and spillover LBI and other channels of learning Manufacturer’s knowledge -0.00043***-0.00037**-0.00035**-0.00029**-0.00035** -0.00029**-0.00025* stock (0.00016)(0.00015)(0.00014)(0.00013)(0.00014) (0.00013) Manufacturer’s experience -0.00488-0.01816-0.00309 -0.00211-0.00216 alone (GW) (0.00489)(0.02285)(0.00515) (0.00572)(0.00571) Developer’s experience in -0.02718***-0.03938*-0.02418*** -0.02333***-0.02230*** CDM projects alone (GW) (0.00456)(0.02017)(0.00611) (0.00658)(0.00655) Cooperating experience in -0.04976** -0.04149**-0.03971** CDM (GW) (0.02037) (0.01874)(0.01876) Spillover from the province -0.000600.00457-0.00027 (GW) (0.00553)(0.00723)(0.00252) Spillover from the industry -0.00767*** -0.00001-0.01605 (GW) (0.00296) (0.00248)(0.02444) Foreign manufacturer* -0.13107** cooperating experience (0.06000) Control variablesYes Province fixed effectsYes Year fixed effectsNoYesNoYesNoYes Observations 486 R-squared0.6040.6680.6740.6840.6600.6850.687

14. Empirical Results 1.Learning by searching 2.Learning by doing and spillover -Wind projects learn from developer’s internal experience in both wind farm installation and operation. -Knowledge spillover effects from the industry as a whole are less important.

15. v Unit Production Costs Note: 1. Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1. 2. Industrial level experience omitted in model (2), (6) and (7) because the sum of different levels of experience equals to the total industry-wide capacity in a given year, which is correlated with the year dummies. Dependent Variable:(1)(2)(3)(4)(5)(6)(7) ln (unit_cost)Aggregate level Internal experience and spillover LBI and other channels of learning Manufacturer’s knowledge -0.00043***-0.00037**-0.00035**-0.00029**-0.00035** -0.00029**-0.00025* stock (0.00016)(0.00015)(0.00014)(0.00013)(0.00014) (0.00013) Manufacturer’s experience -0.00488-0.01816-0.00309 -0.00211-0.00216 alone (GW) (0.00489)(0.02285)(0.00515) (0.00572)(0.00571) Developer’s experience in -0.02718***-0.03938*-0.02418*** -0.02333***-0.02230*** CDM projects alone (GW) (0.00456)(0.02017)(0.00611) (0.00658)(0.00655) Cooperating experience in -0.04976** -0.04149**-0.03971** CDM (GW) (0.02037) (0.01874)(0.01876) Spillover from the province -0.000600.00457-0.00027 (GW) (0.00553)(0.00723)(0.00252) Spillover from the industry -0.00767*** -0.00001-0.01605 (GW) (0.00296) (0.00248)(0.02444) Foreign manufacturer* -0.13107** cooperating experience (0.06000) Control variablesYes Province fixed effectsYes Year fixed effectsNoYesNoYesNoYes Observations 486 R-squared0.6040.6680.6740.6840.6600.6850.687 A typical CDM project by the same developer will lead to around 0.23% to 0.25% decrease in future unit costs.

16. Empirical Results 1.Learning by searching 2.Learning by doing and spillover 3. Cost reduction through repeated collaboration -The repeated partnership between project developer and manufacturer matters, which leads to lower electricity production costs, particularly for capital costs. -However, the cooperating experience does not significantly improve capacity factor.

17. Effects of collaborating experience and technology transfer Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 (1)(2)(3)(4)(5) Dependent Variablesln(unit cost) ln(unit capital cost) ln(capacity factor) Knowledge stock of-0.00029**-0.00025*-0.00031*-0.000270.00030** manufacturer(0.00013) (0.00018)(0.00017)(0.00012) Manufacturer’s experience-0.00211-0.00216-0.00768-0.007720.00512 alone (GW)(0.00572)(0.00571)(0.00704)(0.00703)(0.00556) Developer’s experience in-0.02333***-0.02230***-0.03393***-0.03304***0.01126** CDM projects alone (GW)(0.00658)(0.00655)(0.00800)(0.00798)(0.00573) Cooperating experience in-0.04149**-0.03971**-0.05406**-0.05252**0.00505 CDM (GW)(0.01874)(0.01876)(0.02299)(0.02314)(0.01987) Turbine size (MW)0.010680.009290.03342*0.03221*0.05332*** (0.01481)(0.01493)(0.01817)(0.01825)(0.01093) Project size (GW)-0.34063***-0.33709***-0.34582***-0.34276**0.14699** (0.12296)(0.12246)(0.13293)(0.13291)(0.07374) Wind category 1-0.14092***-0.14541***-0.21581***-0.21969***0.13427*** (0.02805)(0.02850)(0.02739)(0.02766)(0.03026) Wind category 2-0.09387***-0.09348***-0.16919***-0.16885***0.10052*** (0.02688)(0.02706)(0.02500)(0.02479)(0.02982) Wind category 3-0.01251-0.01419-0.02476-0.02621-0.00811 (0.01974)(0.01983)(0.01851)(0.01852)(0.01573) Foreign manufacturer0.03097*0.03969**0.03535*0.04288**0.00991 (0.01607)(0.01751)(0.01951)(0.02138)(0.01422) Central SOE developer-0.01624-0.01554-0.00684-0.006240.00899 (0.01126)(0.01128)(0.01453) (0.01001) Local SOE developer0.03259*0.03183*0.05485**0.05419**-0.01548 (0.01885)(0.01890)(0.02443)(0.02451)(0.01544) Foreign manufacturer* -0.13107** -0.11329 cooperating experience (0.06000) (0.08382) Province fixed effectsYes Year fixed effectsYes Observations486 502 R-squared0.6850.6870.6070.6080.606 One more CDM wind project that a developer builds together with the same foreign manufacturer will reduce the unit cost by almost 1%. -- 4 times of the effects of collaborating with domestic manufacturer

18. Empirical Results 1.Learning by searching 2.Learning by doing and spillover 3.Cost reduction through repeated collaboration 4. Technology transfer through CDM -Learning-by-interacting effect occurs when a wind project developer repeatedly collaborates with a foreign manufacturer. -Transfer of tacit knowledge through partnership

19. Policy Implications For Chinese policymakers: -Increase understanding of the learning process in China’s wind industry -Help to make more targeted policies to forge the partnership between project developers and turbine manufacturers For international climate change policy making: - Shed light on how CDM facilitate technology transfer  by encouraging cooperation between local project developers and foreign turbine manufacturers

20. THANKS! Questions and comments are appreciated.

21. From 2003 to 2010, the cumulative installed capacities in registered CDM projects account for approximately 78.7% of the total installed capacities in China. Background: CDM and China’s Wind Industry

22. Motivation: CDM as a Demand-Side Policy for Wind Technology Project Design Document National Approval Validation Registration Monitoring and Verification Issuance of CERs Project Developer National Authority 1 st Designated Operational Entity (DOE) Executive Board (EB) Project Developer & 2 nd DOE EB Project Development and Construction (1-2 years) Project Development and Construction (1-2 years) Operation (20-25 years) Operation (20-25 years) Project Cycle CDM CycleResponsible Party in each step

23. Data and Empirical Model

24. Descriptive Statistics: Variation of unit costs and capacity factors

25. VariableMeanStd.dev.MaxMinN Unit cost (RMB/kWh)0.5230.0761.0070.335492 Unit capital cost (RMB/kWh)0.4250.0660.7800.256492 Unit O&M cost (RMB/kWh)0.0980.0280.2270.011492 Capacity factor0.2530.0280.3670.144510 Manufacturer's knowledge stock (Decay rate = 0.15)18.4036.9228.80504 Manufacturer's cumulative installed capacity (GW)0.6850.7772.6210504 Developer's cumulative installed capacity in CDM projects (GW)0.7220.9653.3380510 Cooperating installed capacity in CDM projects (GW)0.1230.2140.9380510 Province level cumulative installed capacity (GW)0.7911.0293.6790510 Industrial level cumulative installed capacity (GW)3.5182.4206.6020.056510 Average turbine size (MW)1.340.3830.6510 CDM project size (GW)0.0590.0480.40050.00935510 Foreign manufacturer0.1760.38210510 Central SOE developer0.700.4610510 Local SOE developer0.110.3210510 Private developer0.190.3910510 Wind category 10.20.40010510 Wind category 20.2880.45310510 Wind category 30.1250.33210510 Wind category 40.3870.48710510

26. Background: CDM as a Demand-Side Policy for Wind Technology DomesticInternational Supply Side -National basic research program (973 Program, 1997) -National high-tech R&D program (863 Program, 1986) - National key technology R&D program (TKPs, 1982) Demand Side - National wind concession program (2003- 2008) -Mandatory renewable market share (1997) -Power surcharge for wind power (2006) -Relief of VAT and import tax for wind turbines (2008) Clean Development Mechanisms (CDM)

27. Contributions to the Literature Existing LiteratureThis Research Learning process in wind power Focus on: -Learning through R&D - Learning by doing (Goulder, 2004; Junginger, et al, 2005; Nemet, 2012; Qiu et al, 2012) -Provide empirical evidence on the learning by interacting effect. - Highlight the importance of partnership and collaboration in technological change Technological change in China’s wind industry - Qualitative study - Concentrated on domestic policies -First empirical research on CDM projects - Data improvement on electricity production cost Collaboration Concentrated on public service delivery such as welfare program, health, education etc. Extends empirical study on collaboration to international collaboration on carbon reduction and renewable energy technology diffusion.

28. Effects of Aggregate Level Experience Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1. Industrial level experience omitted in all models due to the multicollinearity. (1)(2)(3) Dependent Variables ln(unit cost)ln(unit capital cost)ln(capacity factor) Manufacturer’s knowledge stock -0.00037**-0.00042**0.00032*** (0.00015)(0.00021)(0.00012) Province level experience (GW) 0.00457-0.00296-0.00875 (0.00723)(0.00888)(0.00595) Turbine size (MW) 0.011490.03683**0.05319*** (0.01503)(0.01858)(0.01097) Project size (GW) -0.32164***-0.33291**0.13045* (0.12414)(0.13833)(0.07282) Wind category 1 -0.13557***-0.20836***0.13308*** (0.02839)(0.02820)(0.03098) Wind category 2 -0.09914***-0.17539***0.10357*** (0.02696)(0.02607)(0.03064) Wind category 3 -0.00948-0.02110-0.00796 (0.02184)(0.02033)(0.01626) Foreign manufacturer 0.03793**0.04489**0.00644 (0.01609)(0.01961)(0.01397) Central SOE developer -0.04186***-0.08436***0.01837** (0.01008)(0.02092)(0.00859) Local SOE developer 0.02352-0.04343*-0.01271 (0.01862)(0.02403)(0.01521) Province fixed effects Yes Year fixed effects Yes Observations 486 502 R-squared 0.6680.5790.603

29. Effects of Developer’s Internal Experience and Spillover Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 Dependent Variables(1)(2)(3) ln(unit cost)ln (unit capital cost)ln(capacity factor) Manufacturer’s knowledge stock -0.00029**-0.00031*0.00028** (0.00013)(0.00018)(0.00012) Manufacturer’ experience (GW) -0.01816-0.020130.00529 (0.02285)(0.02752)(0.00471) Developer’s experience in CDM -0.03938*-0.04638*0.01094** projects (GW) (0.02017)(0.02475)(0.00446) Spillover from the industry (GW) -0.01605-0.012450.00107 (0.02444)(0.02942)(0.00345) Turbine size (MW) 0.010680.03342*0.05349*** (0.01481)(0.01817)(0.01085) Project size (GW) -0.34063***-0.34582***0.14495* (0.12296)(0.13293)(0.07389) Wind category 1 -0.14092***-0.21581***0.13587*** (0.02805)(0.02739)(0.02877) Wind category 2 -0.09387***-0.16919***0.10083*** (0.02688)(0.02500)(0.02837) Wind category 3 -0.01251-0.02476-0.00556 (0.01974)(0.01851)(0.01512) Foreign manufacturer0.03097*0.03535*0.01231 (0.01607)(0.01951)(0.01405) Central SOE developer -0.01624-0.006840.00791 (0.01126)(0.01453)(0.00974) Local SOE developer 0.03259*0.05485**-0.01550 (0.01885)(0.02443)(0.01528) Province fixed effectsYes Year fixed effectsYes Constant-0.45777***-0.64313***-1.62209*** (0.10299)(0.10847)(0.07938) Observations486 502 R-squared0.6840.6070.606 The unit capital cost will fall by 4.64% and the capacity factor will increase by 1.1% if the developer increases additional 1 GW installed capacity in CDM projects.