1. 1 Optimal Technology R&D in the Face of Climate Uncertainty Erin Baker University of Massachusetts, Amherst Presented at Umass INFORMS October 2004

2. 2 Today’s Talk Background on Climate Change How to model R&D programs in top-down models? Theoretical results indicate that  How R&D is modeled matters, and  How increasing risk is modeled matters. Results and insights from numerical model. Including uncertainty in the returns from R&D

3. 3 Climate Change Humans are changing the climate, through the accumulation of greenhouse gasses (GHG). GHG are mainly emitted through the combustion of fossil fuels.

4. 4 Human Contributions to the Greenhouse Effect Carbon Dioxide 55% Nitrous Oxide 6% CFCs 11 and 12 17% Methane 15% Other CFCs 7%

5. 5 Carbon Emissions Due to Fossil Fuel Consumption 1860 -1985

6. 6 Reference Carbon Projections

7. 7 Climate Change We have seen an increase of about 1°C over the last 100 years. A doubling of CO 2 from pre-industrial levels would increase global average temperatures by about 1.5 – 4.5°C A 1.5°C rise would be warmest temperatures in last 6000 years. A 4.5°C rise would raise temperatures to those last seen in time of dinosaurs.

8. 8 Climate Change - Uncertainty 100 years is a blip in geologic time. Climate models are still in infancy. Regional climate impacts are highly uncertain. Human impacts of climate changes are uncertain. Potential for catastrophic damages  Runaway greenhouse effect  Failure of the Gulf stream

9. 9 Climate Change Uncertainty about how emissions today will cause damages tomorrow. But, we are learning more and more. Uncertainty, learning, and adaptation impact current decisions Conclusion: Uncertainty + Learning = less control of emissions.  Kolstad  Ulph & Ulph  Manne & Richels  Baker

10. 10 What about R&D? R&D planning is complicated by different programs  Solar PVs versus efficiency of coal-fired electricity We consider optimal R&D  uncertainty and learning about climate damages  choice of R&D program

11. 11 How to represent R&D? Climate change is a complex problem, involving multiple variables. In order to get insights about the best policy, we need a simple representation of alternative R&D programs. What matters for climate change is how the technology that results from the R&D impacts the cost of abatement.

12. 12 Climate Change Policy We would like to choose a carbon emissions level that equated the marginal cost of abatement with the marginal damages from climate change. MC = MD Technical change impacts the marginal cost of abatement.

13. 13 The Production Function **   0 Q = f(  )  = standard inputs  = emissions

14. 14   Cost **   0 Q = f(  ) 0 $ Production Function Abatement Cost Curve From production function to abatement cost curve  = emission reductions  = emissions

15. 15 Multiplicative Shift: Cost Reduction of No-Carbon Alternatives   max   min 1-    0 01  Production Function Cost The abatement cost curve pivots downward

16. 16      Production Function Cost Emissions Reduction of Currently Economic Alternatives The abatement cost curve pivots to the right

17. 17 Define Risk How does optimal investment in R&D change with an increase in risk? “Risk” – “uncertainty” – “Mean-preserving- spread” See for example Rothschild & Stiglitz 1970,1971.

18. 18 Theory Results Proposition: Optimal R&D decreases with some increases in risk.

19. 19 Theory Results Proposition: Optimal R&D decreases with some increases in risk.  “Full abatement”

20. 20 Theory Results Proposition: Optimal R&D decreases with some increases in risk.  “Full abatement”  Fundamentally different from abatement result

21. 21 Theory Results The converse is not true – some R&D programs will always decrease in risk. Individual R&D programs will react differently to an increase in risk. It is crucial to model the specific program.

22. 22 R&D impacts convexity of cost curve / production function Cost Reduction   Emissions reduction   Flatter  R&D increases in risk More convex  R&D decreases in risk

23. 23 Integrated Assessment Model William Nordhaus’s DICE Optimal Growth + Climate Model  Social Planner chooses how to divide income between consumption, investment, and emissions reduction. Added uncertainty, using stochastic programming.  First 5 periods decisions are made under uncertainty  After 5 periods the world splits into two damage scenarios.

24. 24 Integrated Assessment Model William Nordhaus’s DICE Added R&D as a decision variable.  One time decision in 1 st period before learning  Cost reduction implemented in 50 years, after learning about damages.  No uncertainty in the returns to R&D.

25. 25 Increasing Damage certain low medium high Probability of high damage 0.018.013.002374 Value of high damage -.042.057.3 Value of low damage.0035.002794.002794 002794 Increasing Probabilitycertain low medium high Probability of high damage 0.018.05.08333 Value of high damage -.042.042.042 Value of low damage.0035.002794.001473 0 2 Types of increasing risk

26. Increasing Probability 1 1 0.78 3 0.9 0. 1 0.75 0.25 0.33 3 Damage is on x-axis, Probability is on y-axis Increasing Damage 1 1 0.75 0.25 0.33 3 4 0.82 0.18 0.33

27. 1 1 0.78 3 0.9 0. 1 Damage is on x-axis, Probability is on y-axis 1 1 0.75 0.25 0.33 3 0.67 0.33 0 3 4 0.82 0.18 0.33 Increasing Probability Increasing Damage

28. 1 1 0.78 3 0.9 0. 1 Damage is on x-axis, Probability is on y-axis 0.67 0.33 0 3 5 1 1 0.75 0.25 0.33 3 0.86 0.14 0.33 Increasing Probability Increasing Damage

29. 29 Results – Increasing Probability 0 4 8 12 16 00.020.040.060.080.1 Probability of high damage Billions of US$ Cost ReductionEmissions Reduction 0 0.1 0.2 0.3 0.4 00.020.040.060.080.1 Probability of high damage Optimal R&D

30. 30 Results – Increasing Damages % GDP Loss 0 1 2 3 4 5 020406080 Billions of US $ 0 0.04 0.08 0.12 0.16 020406080 % GDP Loss Optimal R&D Cost ReductionEmissions Reduction

31. 31 Conclusions R&D can be a hedge against uncertainty. But, it depends on what kind of R&D.  R&D into reducing the cost of low carbon alternatives And what kind of risk.  Increasing the probability of needing very low carbon technologies, rather than considering higher levels of damages.

32. 32 Unknowns We need to estimate the relationship between investment in and R&D program, and the expected impact on the abatement cost curve. We need to estimate the amount of uncertainty surrounding R&D programs.

33. 33 Uncertain Returns to R&D

34. 34 DICE equations