IPCC Fourth Assessment Report: Climate Change 2007
Climate Change 2007: Working Group III: Mitigation of Climate Change

2.3.2 Typologies of risk and uncertainty

The literature on risk and uncertainty offers many typologies, often comprising the following classes:

Randomness: risk often refers to situations where there is a well-founded probability distribution in typologies of uncertainty. For example, assuming an unchanged climate, the potential annual supply of wind, sun or hydropower in a given area is only known statistically. In situations of randomness, expected utility maximization is a standard decision-making framework.

Possibility: the degree of ‘not-implausibility’ of a future can be defined rigorously using the notion of acceptable odds, see De Finetti (1937) and Shackle (1949). While it is scientifically controversial to assign a precise probability distribution to a variable in the far distant future determined by social choices such as the global temperature in 2100, some outcomes are not as plausible as others (see the controversy on scenarios in Box 2.2). There are few possibility models related to environmental or energy economics.

Box 2.2 The controversy on quantifying the beliefs in IPCC SRES scenarios

Between its Second and Third Assessment Reports, the Intergovernmental Panel on Climate Change elaborated long-term greenhouse gas emissions scenarios, in part to drive global ocean-atmosphere general circulation models, and ultimately to assess the urgency of action to prevent the risk of climatic change. Using these scenarios led the IPCC to report a range of global warming over the next century from 1.4–5.8°C, without being able to report any likelihood considerations. This range turned out to be controversial, as it dramatically revised the top-range value, which was previously 3.5°C. Yet some combinations of values that lead to high emissions, such as high per-capita income growth and high population growth, appear less likely than other combinations. The debate then fell into the ongoing controversy between the makers and the users of scenarios.

Schneider (2001) and Reilly et al. (2001) argued that the absence of any probability assignment would lead to confusion, as users select arbitrary scenarios or assume equi-probability. As a remedy, Reilly et al. estimated that the 90% confidence limits were 1.1–4.5°C. Using different methods, Wigley and Raper (2001) found 1.7–4.9°C for this 1990 to 2100 warming.

Grübler et al. (2002) and Allen et al. (2001) argued that good scientific arguments preclude determining objective probabilities or the likelihood that future events will occur. They explained why it was the unanimous view of the IPCC report’s lead authors that no method of assigning probabilities to a 100-year climate forecast was sufficiently widely accepted and documented to pass the review process. They underlined the difficulty of assigning reliable probabilities to social and economic trends in the latter half of the 21st century, the difficulty of obtaining consensus range for quintiles such as climate sensitivity, and the possibility of a non-linear geophysical response.

Dessai and Hulme (2004) argued that scenarios could not be meaningfully assigned a probability, except relative to other specific scenarios. While a specific scenario has an infinitesimal probability given the infinity of possible futures, taken as a representative of a cluster of very similar scenarios, it can subjectively be judged more or less likely than another. Nonetheless, a set of scenarios cannot be effectively used to objectively generate a probability distribution for a parameter that is specified in each scenario.

In spite of the difficulty, there is an increasing tendency to estimate probability distribution functions for climate sensitivity, discussed extensively in IPCC (2007a), see Chapter 9, Sections 9.6.2 and 9.6.3 and Chapter 10, Sections 10.5.2 and 10.5.4.

Knightian or Deep Uncertainty: the seminal work by Knight (1921) describes a class of situations where the list of outcomes is known, but the probabilities are imprecise. Under deep uncertainty, reporting a range of plausible values allows decision-makers to apply their own views on precaution. Two families of criteria have been proposed for decision-making in this situation. One family associates a real-valued generalized expected utility to each choice (see Ellesberg, 2001), while the other discards the completeness axiom on the grounds that under deep uncertainty alternative choices may sometimes be incomparable (see Bewley, 2002; Walley, 1991). Results of climate policy analysis under deep uncertainty with imprecise probabilities (Kriegler, 2005; Kriegler et al. 2006) are consistent with the previous findings using classical models.

Structural uncertainty: is characterized by « unknown unknowns ». No model (or discourse) can include all variables and relationships. In energy-economics models, for example, there can easily be structural uncertainty regarding the treatment of the informal sector, market efficiency, or the choice between a Keynesian or a neoclassical view of macro-economic dynamics. Structural uncertainty is attenuated when convergent results are obtained from a variety of different models using different methods, and also when results rely more on direct observations (data) rather than on calculations.

Fuzzyness or vagueness: describes the nature of things that do not fall sharply into one category or another, such as the meaning of ‘sustainable development’ or ‘mitigation costs’. One way to communicate the fuzzyness of the variables determining the ‘Reasons for concern’ about climate change is to use smooth gradients of colours, varying continuously from green to red (see IPCC, 2001a, Figure SPM 2, also known as the ‘burning embers’ diagram). Fuzzy modelling has rarely been used in the climate change mitigation literature so far.

Uncertainty is not only caused by missing information about the state of the world, but also by human volition: global environmental protection is the outcome of social interactions. Not mentioning taboos, psychological and social aspects, these include:

Surprise: which means a discrepancy between a stimulus and pre-established knowledge (Kagan, 2002). Complex systems, both natural and human, exhibit behaviour that was not imagined by observers until it actually happened. By allowing decision-makers to become familiar (in advance) with a number of diverse but plausible futures, scenarios are one way of reducing surprises.

Metaphysical: describes things that are not assigned a truth level because it is generally agreed that they cannot be verified, such as the mysteries of faith, personal tastes or belief systems. Such issues are represented in models by critical parameters, such as discount rates or risk-aversion coefficients. While these parameters cannot be judged to be true or false they can have a bearing on both behaviour and environmental policy-making. Thompson and Raynor (1998) argue that, rather than being obstacles to be overcome, the uneasy coexistence of different conceptions of natural vulnerability and societal fairness is a source of resilience and the key to the institutional plurality that actually enables us to apprehend and adapt to our ever-changing circumstances.

Strategic uncertainty: involves the fact that information is a strategic tool for rational agents. The response to climate change requires coordination at international and national level. Strategic uncertainty is usually formalized with game theory, assuming that one party in a transaction has more (or better) information than the other. The informed party may thus be able to extract a rent from this advantage. Information asymmetry is an important issue for the regulation of firms by governments

and for international agreements. Both adverse selection and moral hazards are key factors in designing efficient mechanisms to mitigate climate change.