Coping with the Uncertainties of Climate Change. Prof. Charles D. Kolstad Stanford University SIEPR, PIE & Economics

Transcription

1 Coping with the Uncertainties of Climate Change Prof. Charles D. Kolstad Stanford University SIEPR, PIE & Economics 1

2 Uncertainty is Complex There are known knowns: there are things we know we know. We also know there are known unknowns; that is to say we know there are some things we do not know. But there are also unknown unknowns the ones we don't know we don't know." Famous contemporary philosopher (2003) 2

3 Dimensions of Uncertainty Uncertainty and risk Some uncertainty quantifiable; some not Uncertainty multifaceted: natural science, damage, costs Uncertainty is different from stochasticity Who is uncertain, who learns and does it matter? Scientists? Regulator? Farmer? Big business? Fat tails As ΔT increases, damage grows more rapidly than π(δt) declines expected damage grows with ΔT Learning Over time, uncertainty changes as we learn; implications for timing of action Some learning can be directed by investing (eg, in R&D) How to direct R&D budgets and incentivize innovation? Measuring the value of information Risk perception People perceive risks differently, generating paradoxes Irreversibilities Some actions are irreversible; uncertainty and learning suggests biasing current action towards precaution Evolution of future technology highly uncertain Endogenous depends on regulatory action adopted Key to costs and damages (though adaptation) Managing risk and uncertainty Decentralized Insurance markets Catastrophe bonds and other derivatives Centralized Government investments in infrastructure Mitigation actions tempered by uncertainty and learning 3

4 Uncertainty and Risk Subjective vs Objective probabilities (Frank Knight) Objective ( Knightian risk ) Probabilities can be assigned to states-of-world Data can be used to generate probabilities Examples: coin flip; probability of rain in Des Moines on May 1, 2013 Subjective ( Knightian uncertainty ) Probabilities less well known Possibly little data on which to base estimates Examples: Probability of West Antarctic Ice Sheet collapsing in 2014 Probability of North Atlantic THC shutdown in 21 st Century 4

5 Uncertainty over Climate Sensitivity (from IPCC AR4,Fig 9.20) What kind of uncertainty is this? CO2 (ppm) Pre-industrial: 280 Current: 390

6 Fat Tails Probability distributions Thin tailed finite support Medium tailed tails drop exponentially (normal distribution) Fat tailed tails drop more slowly The big issue is the likelihood of extreme events (in probability sense) Consider two events (Nordhaus, 2011): Height of American women Mean: 64 ; standard deviation of 3 ; normally distributed Probability of 11 woman (23 sigma): Earthquakes (ie, energy released in earthquake) Seismologists tell us energy released in earthquakes follows fat tailed power law distribution If we had assumed a normal distribution, 2011 Japan quake should occur once every years (using historical data), rather than 10-2 (based on power law distribution). Fat tails and climate change Assume: damage from climate change has fat tail, risk aversion significant THEN expected value of damages is negative infinity Which implies all resources should be devoted to mitigating climate change Implications for climate change policy still a matter for debate 6

7 Learning Learning: time passes, information is acquired. Learning about a parameter may not reduce the variance of your estimate The estimate of modal/mean climate sensitivity largely unchanged over 100+ years (2.5-3 degrees C) Who learns? Individual agents (eg, farmers) who must act in an environment of changing uncertainty and learning Regulators who make decisions about mitigation, adaptation and other actions, in environment of uncertainty and learning Multi-agent decisions (IEAs) How does learning take place? Passively, with the passage of time Active experimentation (update prior on climate sensitivity with one more year of data) Active acquisition of information (R&D) What is the process of active learning? R&D is a fundamental process of information acquisition spend more and learn more (in expectation) Emit more or less GHG will change the strength of the climate signal 7

8 Simple Model of Passive Learning A is prior Truth is B, C or D 8

9 Example: Regulator Learning Regulator uncertain about climate sensitivity Each year provides more data on climate (distribution of weather) Regulator decides on regulatory actions with Uncertainty regarding costs, benefits, consequences Knowing she will know more in a decade One action is deferring action Regulator can also take action to change learning Change climate change R&D budget Promulgate regulations to accelerate learning (eg, fuel economy standards or banning incandescents) 9

10 Example: Farmer Learning Value of ag land relative to US mean Source: Mendelsohn et al,

11 One view of adaptation and adjustment Farmers can instantly change cropping in response to climate change Climate is unobserved: Problem is that farmers do not observe a change in the climate Only after experiencing a change in climate for a number of years will farmers adapt and change Impacts consist of losses after long run adaptation PLUS transient costs ( adjustment costs ) Some action may be precautionary: install irrigation from RISK of climate becoming drier Source: Mendelsohn et al,

12 Annual costs of a change in climate (observed vs. unobserved) Source: Kelly, Kolstad & Mitchell, Adjustment Costs from Environmental Change, J. Env. Econ. & Mgmt., 50: (2005). 12

13 What is the source of these adjustment costs? Weather rarely is average When weather turns out differently than expected, farm output and profits can suffer When the climate has changed and farmer doesn t realize it, farmer suffers more losses After suffering adverse weather for a period of time, farmer realizes climate has changed In the end, everything re-equilibrates 13

14 Empirical implementation Observe annual profits as a function of Farm characteristics Weather expectations (climate from historical record) Actual weather Estimate profit function (of weather and climate) Use panel (time series & cross section) to estimate Data drawn from midwest, Examine counterfactual Unobserved change in climate (consistent with IPCC forecasts) Slow realization that climate has changed (Bayesian farmer) Measure impacts, adjustment costs and adaptation Results for median county Impacts $3.07 per acre gain (5.2% of annual profit) Adjustment costs $1.38 per acre Net impact: $1.69 per acre gain For application, see: Kelly, Kolstad, Mitchell,

15 Perceptions of risk Perceived risk Figure from Kolstad (2011), adapted from Slovic (1979) 15

16 Irreversibilities Examples: Build dam, flood irreplaceably beautiful canyon Increase CO2 concentration in atmosphere Invest in abatement capital to mitigate emissions In each case, action is difficult to quickly reverse Learning key to the issue Take action under uncertainty about state of world Learn true state of the world Try to reverse action under particular states of the world Literature: Abatement equipment irreversibility favors deferring action CO2 in atmosphere favors accelerating action 16

17 Active Learning: R&D Much climate policy hinges on decarbonizing our economy through innovation and invention Two primary mechanisms Direct R&D spending, primarily from federal government Incentives for R&D: carbon taxes, R&D tax incentives, prizes (eg, Golden Carrot), patent protection (IP) Mechanisms and empirical characteristics of these approaches poorly understood R&D spending should be driven by value of information: in which areas would improved knowledge be most valuable Hunch: economics and the costs of mitigation, adaptation and residual damage 17

18 Managing Risk and Uncertainty I. Decentralized Insurance markets Some risks insurable (eg home fire), some not (eg, unemployment) Sometimes institutional change is needed (eg, flood insurance in US vs. Bermuda) Government provision of insurance plagued with difficulties Financial markets Catastrophe bonds (eg, pays $1 if global average temperature rises to x in given year; or if Cat V hurricane hits Gulf in given year) Real options Investing in technology which reduces risk (eg, irrigation for farming insures against drought risk) Decentralization of risk management generally preferred 18

19 Managing Risk and Uncertainty II. Centralized Mitigation Obvious way to reduce risk Best if accompanied by carbon price incentive Adaptation Public investments to ameliorate possible consequences of climate change Institutional reform Subsidized flood insurance tends to magnify damage to US Gulf Coast Water allocation in California tends to magnify water shortage consequences from Climate Change in Calif. International agreements Uncertainty may make agreement easier verdict still out 19

20 Treatment of Uncertainty in Climate Analysis Several types of uncertainty in climate models Parametric uncertainty (parameter has a value but investigator is unsure about it) Structural uncertainty (true model structure is uncertain to investigator) Stochasticity noise that is unknowable ex ante Simplest (and most common analysis) Probability distribution stipulated for key parameters and distribution of outcomes computed Monte Carlo analysis any endogenous decision making implicitly deterministic Structural uncertainty difficult to analyze More difficult (uncommon for complex models) Decision-making under uncertainty endogenous to model Uncertainty changes over time as learning takes place (active or passive learning) In fairness, it is hard enough to treat the problem deterministically; uncertainty is much harder 20

21 Conclusions Risk, uncertainty and learning pervades climate problem Don t look for most likely consequences of a doubling of GHG look at tail events Let the value of information drive further research Focus more on cost of mitigation/adaptation and consequences of a changed climate 21

22 22