Uncertainty, learning and ambiguity in economic models on climate policy: some classical results and new directionsAbstractIntroductionThe basic expected utility frameworkThe effect of uncertainty and learningThe effect of uncertaintyThe effect of learningThe effect of ambiguityIntroduction to the concept of ambiguityThe effect of ambiguity-aversionDiscussion of further applications of ambiguity-aversionSome implications for climate policyAppendixThe effect of uncertainty, learning and ambiguity in the “climate change model”The effect of uncertainty, learning and ambiguity in the “resource depletion model”Partial uncertainty and partial learningThe “irreversibility effect”ReferencesUncertainty, learning and ambiguity in economicmodels on climate policy: some classical resultsand new directionsAndreas Lange & Nicolas TreichReceived: 12 January 2007 / Accepted: 13 December 2007 / Published online: 14 May 2008#Springer Science + Business Media B.V. 2008Abstract We present how uncertainty and learning are classically studied in economicmodels. Specifically, we study a standard expected utility model with two sequentialdecisions, and consider two particular cases of this model to illustrate how uncertainty andlearning may affect climate policy. While uncertainty has generally a negative effect onwelfare, learning has always a positive, and thus opposite, effect. The effects of bothuncertainty and learning on decisions are less clear. Neither uncertainty nor learning can beused as a general argument to increase or reduce emissions today without studying thespecific intertemporal costs and benefits. Considering limits in applying the expected utilityframework to climate change problems, we then consider a more recent framework withambiguity-aversion which accounts for situations of imprecise or multiple probabilitydistributions. We discuss both the impact of ambiguity-aversion on decisions anddifficulties in applying such a non-expected utility framework to a dynamic context.1 IntroductionClimate policy decisions today have to be made under substantial uncertainty: the impact ofaccumulating greenhouse gases in the atmosphere is not perfectly known, the futureeconomic and social consequences of climate change, in particular the valuation of possibledamages, are uncertain. However, learning will change the basis of making future decisionson abatement policies.The issues of uncertainty and learning are often presented in a colloquial sense. Twoopposing effects are typically put forward: First, uncertainty about future climate damage,which is often associated with the possibility of a catastrophic scenario is said to give apremium to slow down global warming and therefore to increase abatement efforts today.Climatic Change (2008) 89:7–21DOI 10.1007/s10584-008-9401-5A. Lange (*)AREC, University of Maryland, 2200 Symons Hall, College Park, MD 20742, USAe-mail: [email protected]. TreichToulouse School of Economics (LERNA-INRA), 21, all. de Brienne, Aile J.-J. Laffont,31000 Toulouse, Francee-mail: [email protected], learning opportunities will reduce scientific uncertainty about climate damage overtime. This is often used as an argument to postpone abatement efforts until new informationis received. The effects of uncertainty and learning on the optimal design of current climatepolicy are still much debated both in the academic and the political arena.In this paper, we present how uncertainty and learning are classically studied ineconomics. The characterization of uncertainty and learning in economics relates to earlyconcepts introduced in mathematics and statistics. We believe that there is an interest inintroducing these formal concepts to an interdisciplinary audience. Indeed, what we presentis now a common and broadly accepted approach in economics to formally study the effectsof uncertainty and learning. Moreover, we illustrate how one can apply this approach togive insights into the climate change problem.We proceed as follows. We first define the concepts of uncertainty and learning withinthe classical framework of economic decision theory, namely the (Bayesian) expectedutility framework. We consider a two-decision model that encompasses most existingmicroeconomic models that have analyzed the effects of uncertainty and learning. For thesake of illustration, we introduce two examples of this model. One example is a “climatechange” model, and the other example is a “resource depletion” model. We show that theattitude of a decision-maker towards risk and the type of payoff function are instrumental tothe sign of the effect of uncertainty and learning on optimal emissions reductions.Specifically, our results indicate that, compared to the reduction of emissions undercertainty, uncertainty and learning generally cannot provide a clear argument for stricterabatement of emissions today or their postponement.While standard economic decision theory relies on an expected utility framework, empiricaland experimental data have long suggested that this framework fails to explain observedindividual choices. In particular, the preferences of individuals in situations of imprecise ormultiple probabilities are often not consistent with a single (objective or subjective) probabilitydistribution as usually assumed by the theory of expected utility. Climate change policy is aclassical example involving imprecise probabilities. Predictions are derived from different modelswhose results are often presented as a range of probabilities for a single event (IPCC 2005).We thus consider an alternative to the expected utility framework that accounts for thistype of uncertainty over probabilities, or “ambiguity”. We thus generalize the previousframework to illustrate some immediate implications of ambiguity for climate policy. Weshow that ambiguity typically leads to stricter abatement policies today. We also point outdifficulties in applying such a non-expected utility theory to a dynamic framework wherebeliefs should be updated frequently to account for new information. We conclude the paperwith a word of caution: the optimal response of climate policy to uncertainty and learning issensitive to which decision theoretical framework is used. Furthermore, alternatives to thestandard expected utility framework may have better descriptive power but also cangenerate unappealing normative effects.2 The basic expected utility frameworkWe first consider a basic decision