Ecosystem Services Economics

The precautionary principle and global environmental change

Terrence Iverson Department of Economics, Colorado State University Charles Perrings ecoSERVICES Group, School of Life Sciences, Arizona State University

December 2009

Papers in this series are not formal publications of UNEP. They are circulated to encourage thought and discussion. The use and citation of this paper should take this into account. The views expressed are those of the authors and should not be attributed to UNEP. Copies are available from the EcosystemServices Economics Unit, Division of Environmental Policy Implementation, UNEP, Nairobi, Kenya. Terrence Iverson acknowledges the contribution of Buz Brock, Stephen Carpenter and Steven Durlauf to the early development of the ideas expressed in this paper.

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

TABLE OF CONTENTS
ABSTRACT 1 INTRODUCTION 2 ELEMENTS OF A PRECAUTIONARY DECISION 2.1 Horizon scanning 2.2 Policy assessment 2.3 Learning 3 DELIBERATION AND DECISION 3.1 Preferences for a high level of protection 4 APPLICATION TO CLIMATE-CHANGE POLICY 5 CONCLUSIONS REFERENCES 3 3 7 8 10 12 13 14 15 21 22

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

ABSTRACT

The precautionary principle is a mandate to tread cautiously when managing novel threats to the environment or human health. A major obstacle when applying the principle at the international level is disagreement about how precautionary efforts should be restricted to ensure that policy costs are proportional to the attained level of protection. Proportionality is an unresolved question when preliminary evidence precludes decision makers from assigning probabilities over future outcomes. The paper suggests practical analytical tools that governments can use to evaluate and communicate feasible options for balancing ex ante trade-offs when probabilities are unavailable. The suggested tools are applied to climate-change policy using the integrated-assessment model DICE (Nordhaus 2008). In addition, the paper situates the task of precautionary decision making within the broader context of implementing a precautionary response at the international level.

KEYWORDS: Precautionary principle, decisions under uncertainty, global environmental change, environmental externality, Multilateral Environmental Agreements, climate change

1 INTRODUCTION
Rapid global change has led to the emergence of novel environmental threats with potentially severe consequences for public health and for the provision of many ecosystem services. Some familiar examples are the threat of dangerous climate change and the risk of emerging infectious diseases (IPCC 2007; Perrings, Fenichel and Kinzig 2010; Jones et al 2008). In each case, there are few historical precedents for the threat, so future impacts are highly uncertain. The effects may also be cumulative and delayed, so an effective response cannot wait for scientific uncertainty to be resolvedpolicy must be anticipatory. Governments have often neglected to act on the basis of early scientific warnings of environmental threats. For example, the evil effects of asbestos dust were first noted by a British medical inspector in 1898, yet asbestos was not banned in the UK until 1998 (UNESCO 2005, Harremos et al 2002). A recurring pattern of regulatory failure has prompted public concern that conventional approaches to risk management may not adequately protect the public interest, particularly when there is potential for severe and irreversible harm. The precautionary principle has emerged as a focal point of this discussion.

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE The precautionary principle is a guideline for regulating novel threats under conditions of severe scientific uncertainty. It is argued to have originated in Germany, in the early 1970s, as the concept of Vorsorgeprinzip (Haigh, 1993)foresight planningwhich emphasizes avoidance of potentially damaging actions even where there is uncertainty about the consequences of those actions. The precautionary principle has since been applied to a wide range of environmental threats. There is, however, little consistency in the way it has been defined. Alternative versions range from the Rio Declaration, a relatively weak assertion that uncertainty should not be used as an excuse to justify inaction, to much stronger assertions. The Wingspread Statement, for example, states: "When an activity raises threats of harm to human health or the environment, precautionary measures should be taken even if some cause-and-effect relationships are not fully established scientifically."

While efforts to establish a universal definition for the precautionary principle have been controversial, mandates to bias decisions toward caution when addressing novel threats have not. For many categories of novel risk, in many countries, a precautionary approach is well established in national law. In many places, the Hippocratic mandate do no harm is reflected in procedures for approving drugs and for responding to the side effects of recently released drugs. The precautionary principle has been adopted in regulatory statutes for health and the environment in Australia, Canada, New Zealand and the UK, and it is a founding component of European environmental law due to its inclusion in the 1992 Maastricht Treaty. In the United States, a precautionary approach is embraced in numerous environmental and consumer statutesthe Delaney clauses to the Food, Drug, and Cosmetic Act prohibit use of food or chemical additives found to be carcinogenic in animals, the Endangered Species Act allows regulators to stop development activity when there is risk of irreversible species extinction, and The Clean Air Act Amendments of 1977 instruct regulators to assess risk rather than wait for proof of actual harm when setting emissions standards (Vogel 2002). At the international level, a precautionary approach was first endorsed in the World Charter for Nature, adopted by the UN general assembly in 1982, while the precautionary principle was first explicitly invoked in the Second International Conference on the Protection of the North Sea (1987). The principle was widely adopted in the 1990 Bergen Declaration (signed by 84 countries as a follow-up to the Bruntland Commission report) and in the 1992 Rio Summit. Versions of the principle have since appeared in a variety of multilateral agreements. Nevertheless, there is little consistency across agreements, and a precautionary approach is not yet a rule of customary international law. A major obstacle to implementing the precautionary principle at the international level stems from the need for multilateral cooperation. Many international threats involve environmental public goods, while the costs of precautionary action entail loss of national benefits. Without a sovereign global authority to mobilize resources, efforts to provide international public goods demand cooperation among nation states. Cooperation is a challenge for decision making when scientific evidence is too preliminary for decision makers to confidently assign probabilitieseither because historical data is limited or because causal mechanisms are not well understood. When probabilities are difficult to assign, formal decision tools for implementing the precautionary principle leave substantial

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

room for stakeholders to disagree. To show this, we briefly review the literature on implementing the precautionary principle as a formal criterion for decision making under uncertainty. We use this discussion to motivate the approach to precautionary decision making that we advocate in the paper. The decision-theoretic literature on implementing the precautionary principle has progressed in two directions. The first emphasizes the dynamic nature of scientific uncertainty. When the consequences of uncertain decisions are irreversible, and when better information is expected in the future, it is optimal for an expected-welfare-maximizing decision maker to bias decisions toward maintaining flexibility. This result is a straightforward application of backward inductionfuture information has value only if the option to use it has been preserved. It implies that the opportunity cost of an irreversible development project should be augmented to include the potential value of future information. If the project were to develop a Redwood grove, future information would include the recreational value of the grove to future generations (Arrow and Fisher 1974). Henry (1974) called this the irreversibility effect. Epstein (1980) provided a more general characterization of the effect, and subsequent work has applied variations of Epsteins result to the question of optimal resource management (for example, Ulf and Ulf 1997). Gollier and Treich (2000, 2003) identify the precautionary principle with the irreversibility effect. They identify an important class of circumstances in which a precautionary bias is justified, and ground the principle in compelling theoretical foundations. However, by tying the precautionary principle to expected utility theory, they make it difficult to apply. When scientific evidence is highly preliminary, it may be difficult for decision makers to assign probabilities across alternative forecasting models, and it is then even more difficult for diverse stakeholders in an international policy decision to agree upon a common probability distribution. Without probabilities, expected utility is undefined. The second direction of the literature identifies the precautionary principle with criteria for decision-making under ambiguity. Ambiguity, as opposed to risk, refers to a situation in which there is no basis for assigning probabilities across alternative scientific models. The criterion for decision making under ambiguity that most clearly captures the notion of conservatism in the face of potentially severe harm is maximin. By design, maximin policies are robust to catastrophethey implement the optimal response to the worst-case scientific model. Maximin attains a high level of protection with respect to environmental threats, but the protection comes at a cost which the criterion ignores. In particular, if the true model turns out to be less dire than the worst-possible case, resources will have been wasted. In practice, when maximin is applied, it is typically the initial criterion in a strategy that evolves over time. As decision-makers gain more confidence in the probabilities attaching to different outcomes, the criterion evolves towards something that looks more like expected utility. Learning and adaptation decrease the potential magnitude of wasted resources. Nevertheless, the opportunity cost of unnecessary protective measures during the learning phase could still be significant. In fact, this concern has been a focal point for critics of the precautionary principle, although most critics emphasize the static version of maximin precaution i.e. precaution without learning (Sunstein 2005).

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE Anticipating this concern, most statements of the precautionary principle recognize the need to constrain policy costs to be proportionalin some senseto the value of protection attained. For example, a recent formulation by UNESCO (2005) states "When human activities may lead to morally unacceptable harm that is scientifically plausible but uncertain, actions shall be taken to avoid or diminish harm. Actions should be chosen that are proportional to the seriousness of the potential harm" (emphasis added). An appealing feature of a criterion that maximizes expected utility is that it balances ex ante costs and ex ante benefits (i.e., expected costs and expected benefits) in a natural and compelling way. When a decision is made under ambiguity, however, the appropriate interpretation of a proportional response is less obvious.

A plausible interpretation is captured by minimax regret, an alternative criterion for decision making under ambiguity. Minimax regret minimizes the worst-case mistake, or regret, as opposed to maximizing welfare under the worst-case model. Regret is the difference between how well a policy does ex post and how well the policy-maker could have done had the true model been known in advance. As we explain in section 3, regret can be decomposed additively into two component mistakes. One is the mistake that environmental damages exceed what would have been optimal from a cost-benefit perspective. The other is the mistake that resources turn out to have been wasted on unnecessary protective measures. Because minimax regret guards in a robust way against overall mistakes, it guards equally against the incidence of each of the two types of mistake (Iverson 2009). In this way, it balances ex ante costs with ex ante benefits and can be interpreted as implementing a proportional response. Lempert and Collins (2007) argue that regret-based robustness rulesclosely related to minimax regretprovide an effective approach for implementing the precautionary principle in settings with severe uncertainty and plausible risk of catastrophe. Some authors contend that minimax regret does not capture the intended meaning of the precautionary principle (Hansson 1997). Whereas minimax regret guards equally against the contrasting dimensions of risk, maximin can be shown to guard unilaterally against the environmental riskit always avoids the mistake of damaging the environmental excessively, while it also always makes the mistake (at least weakly) of wasting resources (Iverson 2009). Critics of minimax-regret precaution argue that precautionary decisions, above all, should guard against the worst possible outcome. Maximin captures this form of conservatism in a very strong way. But there are in fact a continuum of alternatives between maximin and minimax regret that attain a heightened level of protection with respect to severe environmental threats. Any of these alternatives, one could argue, is consistent with a plausible version of the precautionary principle. In this paper, we associate the precautionary principle with criteria for decision making under ambiguity including a parameter to control the relative weight on the environmental mistake (Iverson 2009). By varying the weight from one to infinity, the criterion traces a continuum of policies between minimax regret and maximin. The weight attaching to the environmental outcome may be time varying. The extent to which the environmental objective should be prioritized depends on the value at risk, and the magnitude of trade-offs across the range of scientific models considered. The magnitude of trade-offs, in turn,

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

depends on the expected rate of future learning. We use visual tools to communicate ex ante trade-offs across the constructed range of policy alternatives. The point is not to identify the precautionary principle with a particular value of the weighting parameter, but rather to communicate to the options for trading-off exposure to the different dimensions of risk. The framework provides information that stakeholders could use in a policy setting in which cooperation is important.

2 ELEMENTS OF A PRECAUTIONARY DECISION

Coordinating the actions needed to avoid novel environmental threats is complicated at the international level because appropriate institutions are often lacking (Walker et al 2009). Whatever the institutional framework, however, the decision process involves identification of early warnings, establishment of the case for action, andafter policy has been set adaptation over time as the scientific basis for action changes. Figure 1 situates the four components of a precautionary response. We develop a framework for deliberation and decision in section 4. The remaining components are discussed in the subsections below.

2.1 Horizon scanning

Horizon scanning activities provide the evidence base for anticipatory decisions. Many national governments have offices responsible for horizon scanning. The UK Department for the Environment, Food and Rural Affairs, for example, defines the objectives of its own horizon scanning activities as The systematic examination of potential threats, opportunities and likely future developments which are at the margins of current thinking and planning. Horizon scanning may explore novel and unexpected issues, as well as persistent problems or trends (http://horizonscanning.defra.gov.uk/). But horizon scanning is a much more widespread dimension of strategic planning at all levels. In many cases it is based on breaking scientific results. Indeed, it may be interpreted as the activity by which the scientific community draws attention to results that potentially have significant consequences for human well-being, and that warrant actionwhether research or policy. Horizon scanning may be a precursor to targeted assessments, research initiatives, policy initiatives or to either private or public defensive responses. In the former US Office of Technology Assessment (OTA) it was linked to the preparation of targeted assessments. More generally, however, it is the basis for determining what forward action to take in cases where there is both novelty and uncertainty. At the international level there are few examples of horizon scanning activities. However, a number of intergovernmental organizations, international non-governmental organizations and multilateral environmental agreements have horizon scanning functions. In addition, some of the global assessment bodies, such as

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE the Intergovernmental Panel on Climate Change and the Millennium Ecosystem Assessment, include a forward-looking component.

Figure 1: Elements of a precautionary decision. The proposed Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services, currently under discussion by the global community, may also have a horizonscanning element. If approved, its functions would be expected to include the development, in association with the global change programs, of a scientific program of research to better understand and predict the causes and consequences of changes in the biosphere at the global scale; regular independent monitoring and assessment of changes in the biosphere at multiple scales; and a responsibility to proactively alert the international community to emerging issues and threats in order to allow timely responses. For the information generated through horizon scanning to trigger a precautionary response, a number of attributes should be present: It should not support conventional risk analysis It should indicate the potential for severe and irreversible harm to the environment or human healthan outcome deemed unacceptable (UNESCO 2005) It should be scientifically plausible (box 1 discusses some qualifying criteria for early warnings)

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

Box 1. Early Warnings Extreme, unique, rare and irreversible events all have either no or only a few historical precedents. It may not be possible to predict a probability distribution of outcomes with confidence. If decision-makers are to act they need sufficient information. That is, they need to be able to recognize and respond to early warning signs. Harremos et al (2001) showed how governments have assessed the likelihood of uncertain outcomes that might include irreversible, significant and widespread effects. This implies a weaker test than that required by standard scientific proof. Historic examples of such tests include scientifically based suspicion, reasonable grounds for concern and the balance of evidence. All are based on data that fail standard tests of scientific proof, but nevertheless provide a basis for decisionmaking. They are early warning signs rather than conclusive proof. The source of scientifically based suspicion or reasonable grounds for concern is frequently a model of the system that is used to simulate possible futures given current conditions. Indeed model predictions provide the most important early warnings. The scenarios familiar from the Intergovernmental Panel on Climate Change, and now from the Millennium Assessment, are examples of model predictions. Modeling offers a transparent basis for predictions. If preliminary theoretical research indicates the possibility of outcomes that involve significant irreversible cost, then the precautionary response involves actions that avoid those costs while using data to test the model predictions. The early warnings described in Harremos et al (2001) are all in the form of isolated experiments or events. They did not therefore provide a basis for estimating a probability distribution of outcomes. A common feature of the case studies reviewed in the report is their novelty. In the case of halocarbons, polychlorinated biphenyls (PCBs) and methyl tertbutyl ether (MTBE), the report argues that novelty, plus the persistence of the chemicals in the environment and their dispersion rate should have indicated a potential problem. Indeed, evidence of novelty, persistence and bioaccumulation are argued to provide a sufficient early warning to trigger precautionary action. An equally common feature of the cases reviewed is that early experimental results did not in fact trigger action. The results of animal tests of PCBs were available in 1937, for example, but were ignored. The contamination of the Great Lakes illustrates the point. Once it had been shown that the lakes contained organochlorine pesticides in the 1960s, research followed showing the bioaccumulation in birds of dichlorodiphenyltrichloroethane (DDT) and PCBs. The accumulation of evidence against DDT led to its banning both in the US and Canada by the early 1970s. But action against PCBs has been both slower and weaker. A series of studies of the human development impacts of exposure to PCBs showed significant development deficits, but the evidence was still regarded as too weak to justify precautionary action. Current levels of PCBs in the Great Lakes are still two orders of magnitude higher than those thought to pose a significant cancer risk in humans (Gilbertson, 2001). It follows that alongside grounds for action, a precautionary decision requires a legal and an institutional framework that allows action to be taken. In many cases the scientific community may be persuaded that hypothesized outcomes are realistic, but there is no responsible authority to take precautionary action.

10

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE In addition, the case for precaution is strengthened when learning is expected to resolve ambiguity quickly, when effects would be widespread or the outlook for future generations would be irreversibly diminished. Reasons why a precautionary response might not be triggered include the determination that conventional risk analysis is adequate, or that the potential harm is not severe enough or plausible enough to justify current action. In the latter case, instituting mechanisms for learning and review may be an appropriate initial response that could lead to precautionary actions later.

2.2 Policy assessment

The policy assessment stage develops the inputs for decision analysis. They entail a description of scientific uncertainty, estimates for costs and benefits, and a forecast of anticipated prospects for learning. If scientific uncertainty can be quantified using probabilistic statements, then conventional risk analysis methods should be used. But if evidence and knowledge are insufficient for probabilities to be plausibly specifiedas assumed in the remainder of this paperthen policy choice can be viewed as a decision under ambiguity. To proceed, a set of qualifying scientific perspectives (models) must first be identified. Models themselves may be probabilistic even though ambiguity is assumed to hold across model alternatives. The qualifying set of models is constructed by identifying those that satisfy a specified threshold of plausibility. This determines the extent to which minority scientific views are admitted into consideration and should balance scientific plausibility with the need to respond to potential early warnings (see box 1 for a description of criteria for assessing plausibility). The next step is to translate environmental effects and policy costs into commensurate values. This task raises a number of familiar issues in cost-benefit analysis. One difficult but essential challenge is to ensure that valuations reflect all direct and indirect impacts, whether reflected in market prices or not. Where novel change affects international public goods, for example, this requires assessment of impacts from a global perspective. Assessment of national impacts alone will underestimate the full collective benefits of the international public good. This may occur because the mandate of the treaty is too narrow or because the threat is not viewed with sufficient urgency to instigate collective action at a truly global scale. Box 2 illustrates this concern in the context of existing treaties for managing invasive pests and pathogens. Finally, the time needed for ambiguity to be resolvedor the expected rate of learning determines the period over which precautionary policies restrict activities. This significantly affects the expected cost of policy. For threats like climate change, where policy analysis must forecast an evolving complex system and relevant outcomes unfold over centuries, severe uncertainty is likely to persist for decades (Roe and Baker 2007). In contrast, for new infectious diseases, like SARS or Swine Flu, ambiguity may be resolved in a matter of weeks.

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

11

Box 2. Advantage of a global perspective in benefit evaluation Consider a specific problemthe dispersal of pests and pathogens through the growth of trade and travel. This is a problem that involves potentially very significant harm, as in the case of the global spread of some highly virulent disease, but about which there is considerable ambiguity. It is not possible to say where emergent infectious diseases are likely to occur, or what their impact is likely to be. It is also a problem that is global in reach and that requires international cooperation. The control of the spread of harmful pests and pathogens is a public good. Once provided, the benefits it offers in terms of enhanced protection of human, animal and plant health and the productivity of agriculture, forestry, aquaculture and fisheries, are available to everyone (Perrings et al, 2002). Like all public goods it will be undersupplied if left to the market. This makes it a collective responsibilitya legitimate role of government at many different scales. Cooperation between nation states on this problem occurs within the framework provided by two main multilateral agreements: the International Health Regulations (IHR) and the Sanitary and Phytosanitary (SPS) Agreement of the General Agreement on Tariffs and Trade (GATT). Although there is a precautionary element in each agreement, it is markedly different (Perrings et al, 2010). The 2005 IHR administered by the World Health Organization (WHO) mandates coordinated and cooperative international action to address emerging human health risks. This includes action to isolate areas thought to be the source of novel risks to give scientists the time needed to understand the extent of those risks and governments the time to put defensive strategies in place. At the same time, the regulations require member states to take account of the opportunity cost of these restrictions. Specifically, the IHR aims to prevent, protect against, control and provide a public health response to the international spread of disease in ways that are commensurate with and restricted to public health risks, and which avoid unnecessary interference with international traffic and trade (IHR, 2005: Article 2). By contrast, the SPS Agreementwhich addresses the risks posed by the spread of pests and pathogens to nonhuman speciesfocuses exclusively on risks to individual countries. In doing so, it emphasizes the avoidance of damage through trade restrictions above risks to the global community. By way of protection, the SPS Agreement authorizes countries to undertake defensive actions (inspections, quarantine, exclusions) in response to the risks posed by pests and pathogens that may be introduced through international trade. Under the World Trade Organization (WTO) and its constituent agreements, particularly the General Agreement on Tariffs and Trade and the SPS Agreement, countries have the right to take actions to protect food safety, animal or plant health. But there is no mechanism to address the collective risks posed by the international trading system other than the successive renegotiations of the GATT conducted by the World Trade Organization (WTO). The aims of the SPS Agreement are to ensure that the defensive measures taken by members to protect their own human, animal and plant health do not constitute a barrier to trade (SPS Agreement, 1995: Article 2). The emphasis on the costs of trade-restrictions and the restriction of the worst case to national damages prevents nations from responding to global threats in a precautionary way. In contrast, the IHR does support global precautionary action (Perrings et al, 2010).

12

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

2.3 Learning
The last component of a precautionary approach is a mechanism for learning. Since uncertainty implies that there is scope for learning (although learning does not necessarily reduce uncertainty), every problem that is a candidate for a precautionary approach is also a candidate for learning. A precautionary decision-making process requires a mechanism for updating the information on which decisions are taken. Indeed, the defining characteristic of a precautionary approach is that it enables learning without risking the most severe (but uncertain) consequences of an action. Since precautionary action is triggered by the potentially severe costs of novel policies, activities or events, learning should establish whether an action does not involve such costs. The important question to ask of precautionary processes, therefore, is what incentive people undertaking activities with uncertain outcomes have to inform themselves of those outcomes. Harremos et al (2001) identify both institutional and disciplinary barriers. Particular problems may be captured by particular disciplines or institutions, with the result that important aspects of the problem are simply ignored. Examples include asbestos and ionising radiation, captured by medical clinicians who focused only on the immediate acute effects of exposure, or bovine spongiform encepalopathy (BSE), captured by veterinarians who regarded the risk of transmission to humans as acceptably slight. But the incentive to learn about the consequences of novel activities is also closely related to the burden of proof. If someone undertakes an activity thought to have harmful effects on others, the legal burden of proof generally rests with those who maintain that the effect exists. The legal burden of proof does not rest with those who maintain that the effect does not exist. The link between smoking and lung cancer is a good example of this, but most environmental effects come into the same category. The burden of proof rests with society, rather than with those whose activities are the source of potential damage. As with grounds for action, therefore, it follows that learning also depends on the establishment of an appropriate institutional and legal structure.

3 DELIBERATION AND DECISION

To clarify the options facing decision-makers, we formulate the policy problem as an expert panel problem (Woodward and Bishop, 1997). Policy-makers face a panel of experts, each expert prescribes an alternative scientific model, and policy-makers have no basis for assigning probabilities across model alternatives. This gives rise to a decision problem under ambiguity. With no single-best criterion for decision-making under ambiguity, a crucial question is: How and to what extent should the attained level of environmental protection be balanced with the ex ante risk of wasting resources? What is meant by a precautionary and proportional response?

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

13

3.1 Trading off environmental harm and the costs of precaution

There are two types of mistakes that policy-makers are liable to make when setting precautionary policy under uncertainty. The mistakes align with the dominant concern of each side in the typical environmental policy debate. One is to set policy that doesnt do enough, with environmental (health) damages exceeding what would have been optimal in a cost-benefit sense. The other is to waste resources on unnecessary protective measures. Efforts to reduce exposure to the incidence of each type of mistake are in conflict since the more resources policy-makers commit to protect the environment (health), thus reducing exposure to the environmental (health) mistake, the more they increase the chance that resources will turn out to be wastedand vice versa. As a result, there are trade-offs between these two types of mistake. It is accordingly helpful to think of a decision criterion with a weighting parameter that controls the relative priority attached to the competing objectives of reducing exposure to the incidence of each type of mistake (Iverson 2009). This can be done by constructing a modified version of minimax regret. Minimax regret guards against high occurrences of regret, or overall ex post mistakes. Regret, however, can be decomposed into two additive components, which correspond to the two types of mistakes that policy-makers are liable to make. Using this decomposition, a version of minimax regret can be constructed that allows decision makers to guard disproportionately against the environmental mistake**. When the weight is set to one, the criterion is equivalent to minimax regret; it then guards against the two mistakes equally. As the weighting parameter increases, the generated policies become more precautionary as the environmental objective takes precedence over the risk of wasting resources. As the weight becomes large, the generated policies converge to maximin, which guards against the environmental mistake unilaterally. The social objective implemented under intermediate weights on the environmental mistake is consistent with a line of reasoning used by the European Commission to motivate precautionary thinking. The report Late Lessons from Early Warnings (Harremos et al, 2001) was generated for the European Commission to review the evidence on precaution. It describes an extensive collection of historical cases in which early warnings indicated potentially serious threats to public health or the environment, but preventative measures were delayed or avoided. The report emphasizes that policy-makers should "focus on the potential costs of being wrong." The mistake that seems unconscionable in hindsight entails false negativesinstances where severe harm to the environment or human health could have been avoided with relative ease if early warnings had been heeded, but they weren't. A criterion that places disproportionate weight on the environmental mistake avoids this type of mistake, but at the cost of wasting resources.

Athirdmistakewouldbetochooseapolicythatisdominatedinthesensethatitalwaysdoesworse thananotherfeasiblepolicy,undereverypossibleoutcome.Weassumeherethatdominatedpolicies havebeenruledout. ** Onecouldalsousethecriteriontoguarddisproportionatelyagainstthemistakeofwastingresources.

14

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE Our goal is not to identify the precautionary principle with a particular value of the weighting parameter, but to communicate the options for trading-off exposure to the different risks. To make the trade-offs between different risks explicit, we employ visual tools that display the magnitude of each type of mistake associated with a given policy, across the range of plausible models (Iverson 2009). The tools are demonstrated in section 4. The magnitude of trade-offs reflects the relative size of policy costs vis-a-vis the worstcase environmental harm and the expected rate of learning.

Note that the weight attaching to the environmental outcome may be time varying. The most common examples of precaution in the protection of public healthproduct recallscan be interpreted as decisions in which a novel outcome (e.g. product failure or health response) triggers a reaction that is initially highly precautionary, but that is relaxed as information on the true likelihood of the outcome emerges. Indeed, the argument for prioritizing risks to the environment or human health is strongest when policies are viewed as temporary measuresproviding an opportunity for learning to occur and preserving flexibility so future policy can adapt. Without learning and adaptation, maximin is hard to justify because it entails a significant risk that resources will turn out to be wasted. However, when learning is expected to unfold rapidly and there is risk of severe and irreversible harm, maximin may provide a sensible rule of thumb for a time-limited response. Recent examples of temporary, maximin responses occurring under the International Health Regulations are the closure of Toronto in response to SARS and the movement restrictions in response to Swine Flu. Of course, maximin is just one option from a range of alternatives. The extent to which the environmental objective should be prioritized depends on the identified case for precaution and the magnitude of trade-offs across the panel of expert models. Ultimately, policy choice is a matter of collective judgment that should combine information about ex ante cost-benefit trade-offs with ancillary considerations that society chooses to value in the context of a specific application. The suggested framework provides useful information to facilitate this process of deliberation.

3.2 Preferences for a high level of protection

The decision to choose a high level of protection with respect to specific risks hinges on the preferences of the decision-maker (or the people they represent). Empirical evidence on choice behavior under uncertainty suggests at least two explanations for cautious behavior. One explanation is based on loss aversion, the tendency for people to weight "relative losses" more acutely than "relative gains". Loss aversion is a component of Prospect Theory, a generalization of expected utility theory constructed to match empirical evidence that uncertain choice behavior is highly sensitive to the reference point of the decision maker (Kahneman and Tversky 1979). Applied to long-term policy decisions like climate change policy, loss aversion suggests that people would tend to weight the risk of damaging the environment more heavily than the competing risk that resources might be wasted.

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

15

The argument presumes that the reference point of the decision maker setting climate policy includes the current state of the climate system and the current level of consumption. Relative to this reference point, a damaged future climate system is in a diminished state, so climate damages entail a relative loss. In contrast, changes in consumption due to unnecessary abatement costs occur, in most scenarios, on the back of expected overall consumption growth. Hence, the level of future consumption, while lower than it could have been, is still higher than today, so the resulting loss appears as a diminished gain, not a relative loss. In experimental studies, people weight relative losses about 2.25 times more than they do relative gains. A second explanation is based on peoples tendency to overweight low-probability, highconsequence events relative to what standard economic theoryexpected utility theory would suggest as rational (McDaniels et al., 1992). As a result, potentially catastrophic or dreadful outcomes may receive disproportionate attention in the policy process, leading to a preference for precaution. Whether or not such fears are warranted by the data, they have an important place in collective decision-making, and should be taken seriously. At the same time, where precaution is triggered by initially high weights accorded to the worst case, it is important to generate the information needed for people to adjust the weight they assign to the worst case.

4 APPLICATION TO CLIMATE POLICY

Climate change is a global threat that satisfies many of the conditions for which a precautionary approach is indicated. The global climate system is immensely complex and relevant data is limited, making it extremely difficult to assign probabilities across alternative forecasting models (Manning 2006). In addition, damages are potentially irreversible and severe. Carbon dioxide emissions are irreversible on a time frame of less than 100 years, and a variety of irreversible environmental effects are possible, including a shift in the Gulf Stream and species extinctions. Even without environmental irreversibilities, the damages associated with climate change are expected to be severe. Nordhaus (2008), with conservative assumptions, estimates that unabated climate change would lead to a 2.5 percent loss in global output by the end of this century. These effects are widespread and public, with future generations bearing the brunt of damages. Use of a precautionary approach is also indicated by the terms of the UN Framework Convention on Climate Change. To illustrate a precautionary approach, we employ Nordhauss (2008) Dynamic Integrated model of Climate and the Economy (DICE). The application builds on work in Iverson (2009). DICE couples a neoclassical model of the world economy with reduced-form descriptions of the carbon cycle and global climate system. Disagreement among scientific experts is captured by uncertainty in climate sensitivity. Climate sensitivity is the equilibrium, global-mean temperature change that would result from a doubled concentration of carbon dioxide relative to its pre-industrial level. It is a summary measure of the overall, long-run effect of rising greenhouse gases on global temperature. Climate sensitivity

16

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE depends on complex feedback mechanisms that have been the subject of considerable scientific debate (IPCC 2007). The IPCC Fourth Assessment Report (IPCC 2007) describes the plausible range of climate sensitivity to be between 2.0 and 4.5 degrees Celsius. We use this range to discipline the range of plausible disagreement.

Damages and policy costs in DICE are set to match the broad scientific literature on climatepolicy costs and benefits. The discount rate is calibrated to match historical data on the real return on capitalabout 4%. This approach is common in the economics literature, but is controversial in applications to climate policy because damages are concentrated in the distant future. At a discount rate of 4%, $1000 worth of damages occurring 100 years from now is worth $20 today. Recent examples of climate policy analyses that employ lower discount rates are Stern 2007 and Heal 2008. The model also does not explicitly account for the possibility of severe threshold events like the potential collapse of the Thermohaline Circulation in the North Atlantic. An augmented version of DICE that includes a probabilistic threshold for catastrophe is Keller et al (2004). With the set of plausible scientific models and relationships for costs and benefits in place, the decision tools suggested in section 3 are applied under two possible scenarios for future learning. The scenarios are illustrative, and the exercise could of course be repeated with different assumptions. Figures 2 and 3 show asymmetric-minimax-regret policy alternatives and the corresponding ex ante trade-offs constructed under a scenario of no learning policies chosen today, amid current ambiguity, are assumed to remain in place forever. Figures 4 and 5 repeat the exercise assuming that the true scientific model will be learned perfectly in 2075, with policies adjusted optimally thereafter. As expected, the magnitude of mistakes are significantly smaller when future policy adjusts in response to future learning. Nevertheless, the policies chosen by the asymmetric minimax regret criterion turn out to be nearly the same under each scenario. Figure 2 displays abatement policies generated under a range of values for the relative weight on the environmental mistake (E). Decades are shown on the horizontal axis and "fraction of global emissions to be abated" is depicted on the vertical axis. The top solid line is the policy generated by maximinit is the optimal policy given temperature sensitivity of 4.5 degrees C. The bottom solid line is the policy generated by maximaxit is the optimal policy given temperature sensitivity of 2.0 degrees C. These extremes display the range of policies that can be justified, on cost-benefit terms, given the range of expert disagreement. The middle solid line is the policy generated by minimax regret, and the dashed lines indicate intermediate policies that weight the environmental mistake more than the mistake of wasting resources. The weighting parameters of 1.3 and 2.2 were chosen to generate equally-spaced policies. By varying the weighting parameter, the asymmetric criterion traces out a continuum of policies that converge to maximin as the weight becomes large.

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

17

Figure 1: Abatement paths for a range of asymmetric minimax regret policies. Taking the four precautionary policies in figure 1, figure 2 displays the corresponding exposure to the two dimensions of risk. Mistakes are shown on the vertical axes in trillions of 2005 US dollars. White bars denote environmental mistakes; black bars denote growth mistakes. A negative environmental mistake means that the chosen policy overcompensated, protecting the environment more than what would have been cost-effective in hindsight. A negative growth mistake means that the chosen policy spent less on protection than would have been optimal in hindsight. Each panel displays the incidence of the two mistakes across a representative sample of models from the plausible range. The first panel in figure 2 demonstrates the sense in which maximin guards unilaterally against the environmental concern: it always avoids the environmental mistake (white bars are never positive), while it always makes the mistakeat least weaklyof having wasted resources (black bars are never negative). According to the exercise, if the maximin policy is kept in place from now into the distant future, and if the true value for climate sensitivity turns out to be 2.0 degrees Celsius, then the world community will have spent, in present value terms, about 2.5 trillion dollars more on greenhouse gas abatement then it would have under the optimal policy. At the same time, it will have reduced climate damages by about 1.5 trillion dollars more than it would have under the optimal policyagain in present value terms. As a point of comparison, world GDP in 2008, calculated at purchasing power parity exchange rates, was about 70 trillion dollars.

18

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

Figure 2: Component mistakes on vertical axes in trillions of 2005 US dollars; climate sensitivity on horizontal axes. White bars indicate environmental mistakes; black bars indicate growth mistakes. Each panel presents the profile of risks for the indicated policy from figure 2. Moving to the right in figure 2 through less aggressive policiesstepping down the ladder of policies in figure 1the black bars get pushed down and the white bars get pushed up; less aggressive policies reduce the incidence of the growth mistake, while increasing the incidence of the environmental mistake. This movement reflects the trade-offs at the heart of the policy decision. Higher levels of environmental protection come at a costrepresented in ex ante terms as higher exposure to the incidence of the mistake that resources could turn out to have been wasted.

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

19

Figure 4: Black dots indicate the MR policy without learning. The grey fork indicates the minimax regret policy with perfect learning in 2075. The upper branch is the optimal continuation under low damages; the lower branch is the optimal continuation under high damages Next, we repeat the analysis under the assumption that the true model is revealed in 2075, with polices assumed to adapt optimally thereafter. Figure 4 compares the generated policies under the learning and no-learning assumptions when the decision criterion is minimax regret. The dotted line denotes the minimax regret policy without learning; it is the same policy seen in figure 2. The solid line indicates the minimax regret policy with perfect learning in 2075. The learning policy forks in the period in which the true model is learned: the upper branch indicates the optimal continuation policy when the true model turns out to have temperature sensitivity of 2.0 degrees C; the lower branch is the optimal continuation when the true model turns out to have temperature sensitivity of 4.5 degrees C. Intermediate values of temperature sensitivity are not shown because they do not affect the policy chosen by the decision criterion. The figure shows that the short-run policy under minimax regret is largely insensitive to the assumption about learning: minimax regret picks nearly the same policy for the periods leading up to 2075 whether learning is expected or not. This means that the irreversibility effect is insignificant for this application when the decision criterion is minimax regret. Similar results are found for the other decision criteria considered in the paper.

20

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

Figure 5: Component mistakes on vertical axes in trillions of 2005 US dollars; climate sensitivity on horizontal axes. White bars indicate environmental mistakes; black bars indicate growth mistakes. Policies and regrets assume perfect learning in 2075. Figure 5 displays ex ante trade-offs under the assumption of perfect learning in 2075. The vertical scale is the same as in figure 3. As one would expect, the ability to adjust policy in response to future learning brings the realized policy trajectory closer to the ex post, modelspecific optima, thus decreasing the magnitude of both mistakes. The magnitude of mistakes under the assumption of perfect learning in 2075 is roughly half that without learning. If one expects climate science to reduce uncertainties significantly within the next centuryand for global environmental policy to adjust nimbly in responsethen the trade-offs displayed in figure 5 are more plausible. As an interesting point of comparison, the uncertainty range for climate sensitivity has not changed substantially in 30 years despite hundreds of published studies attempting to estimate it (Roe-Baker 2007). The point is not to identify the precautionary principle with a particular value of the weighting parameter. Rather, it is to communicate ex ante trade-offs associated with feasible options for balancing exposure to the competing dimensions of risk. When probabilities are available, expected costs and expected benefits entail relevant ex ante measures upon which optimal policies can be based. Without probabilities, visual displays of exposure to the incidence of the competing mistakes provide an alternate way to assess ex ante trade-offs.

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

21

5 CONCLUSIONS
There is pressing need for international decision makers to anticipate and adequately respond to novel risks to the environment and public health. The precautionary principle is an important guideline for setting anticipatory policy on the basis of incomplete scientific knowledge. Yet it has been controversial, in part because of disagreement about how to implement it when scientific understanding of a novel threat is preliminary. An important source of disagreement is the question of proportionalitythe extent to which the ex ante consideration of costs should be admitted alongside the competing desire to achieve a high level of protection with respect to identified threat to the environment (or public health). Proportionality is an unresolved question when preliminary evidence precludes use of probabilistic methods for assessing risk. This issue becomes especially important for international decisions where transnational cooperation is important. There are, however, practical analytical tools that governments can use to evaluate and communicate feasible options for balancing ex ante trade-offs. The analysis shows the sense in which maximinthe criterion most commonly associated with the precautionary principleis insensitive to trade-offs. Maximins conservatism may be unreasonable in settings, like climate policy, in which the relative cost of preventive measures are high and the rate of learning is slow. In policy applications where the cost of prevention is relatively low, where irreversible harm is possible, and where learning is expected to proceed rapidly, maximin may offer a reasonable time-limited response. The analysis considers an array of less-conservative alternatives. Minimax regret weights false negatives on par with false positives, while the intermediate alternatives weight false negatives more. Guarding disproportionately, but not unilaterally, against false negatives implements a version of precaution that balances caution with the desire for a proportional response. The appropriate balance will likely depend on context; it is a matter of collective judgment that should incorporate ex ante trade-offs alongside ancillary considerations that decision makers choose to value. The problem is exacerbated at the international level by the fact that governments have to agree to the weighting function that establishes at least the initial value given to the protection of environmental or public health. The approach sketched in this paper provides a way of communicating the environmental and resource risks associated with differing levels of precaution. It therefore makes it possible for governments to evaluate the opportunity cost of a more or less precautionary approach, and to negotiate accordingly.

22

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

REFERENCES
Arrow, K. and A. Fisher 1974. Environmental Preservation, Uncertainty and Irreversibility, Quarterly Journal of Economics 88, 312-319. Editorial. (2000). Fear of the Future, Wall Street Journal. Published on February 10, 2000. Epstein, L. 1980. Decision-Making and Temporal Resolution of Uncertainty, International Economic Review 21, 269-284. Haigh N. 1993. The precautionary principle in British Environmental Policy. Institute for European Environmental Policy, London. Gilbertson M. 2001. The precautionary principle and early warnings of chemical contamination of the Great Lakes, in Harremos P. Gee D., MacGarvin M., Stirling A., Keys J., Wynne B. and Guedes Vaz S., eds, 2001. Late Lessons from Early Warnings: The Precautionary Principle, Environment Issue Report No 22, European Environment Agency, Copenhagen: 126-134. Gollier, C., J. Bruno, N. Treich 2000. Scientific Progress and Irreversibility: An Economic Interpretation of the Precautionary Principle, Journal of Public Economics 75, 229-253. Gollier, C., N. Treich 2003. Decision-Making Under Scientific Uncertainty: The Economics of the Precautionary Principle, The Journal of Risk and Uncertainty, 27(1):77103. Harremos, P. D. Gee, J. Keys, M. MacGarvin, A. Stirling, S. Vaz, B. Wynne, editors. Late Lessons from Early Warning: the Precautionary Principle 1896-2000. Copenhagen: European Environment Agency; 2001. Hansson, S. 1997. The Limits of Precaution, Foundations of Science 2(2):293-306. IPCC (2004): Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change.Cambridge University Press. Iverson T. 2009. Decision Support for Climate Change Policy. Ph.D. dissertation, University of Wisconsin-Madison. Jones, K., N. Patel, M. Levy, A. Storeygard, D. Balk, J. Gittleman, P. Daszak 2008. Global Trends in Emerging Infectious Diseases. Nature 451(7181): 990-993. Kahneman D. Slovic P. Tversky A. eds. 1982. Judgement under Uncertainty: Heuristic and Biases, Cambridge University Press, Cambridge. Keller, K., B. Bolker, and D. Bradford. 2004. Uncertain Climate Thresholds and Optimal Economic Growth. Journal of Environmental Economicsand Management 48(1):723741. Knight, F. H. (1921). Risk, Uncertainty, and Profit. Boston: Houghton Mifflin.

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

23

Lempert, R. and M. Collins 2007. Managing the Risk of Uncertain Threshold Responses: Comparison of Robust, Optimum, and Precautionary Approaches, Risk Analysis 37(4): 1009-1026.
Ludwig, D., W. Brock and S. Carpenter 2005.Uncertainty in discount models and environmental accounting, Ecology and Society 10(2): 13. Manning, M. 2006. The treatment of uncertainties in the Fourth IPCC Assessment Report. Advances in Climate Change Research 2(1):1321.

Manski, C. 2008.Identification for Prediction and Decision.Harvard University Press. McDaniels T, Kamlet, M. and Fischer, G. 1992. Risk perception and the value of safety. Risk Analysis 12(4):495-503. Nordhaus, W. 2008.A Question of Balance: Weighing the Options on Global Warming Policies. Yale University Press. Perrings, C. 1991. Reserved Rationality and the Precautionary Principle:Technological Change, Time, and Uncertainty in Environmental Decision-Making. Ecological Economics: the Science and Management of Sustainability, 153166. Perrings C., E. Fenichel and A. Kinzig 2010. Externalities of globalization: bioinvasions and trade. pp 42-55 in Perrings, C., H. Mooney and M. Williamson (eds) Globalization and Bioinvasions: Ecology, Economics, Management and Policy, Oxford, Oxford University Press. Perrings C., H. Mooney, M. Lonsdale and S. Burgeil 2010. Globalization and invasive species: Policy and management options. pp 235-249 in Perrings, C., H. Mooney and M. Williamson (eds) Globalization and Bioinvasions: Ecology, Economics, Management and Policy, Oxford, Oxford University Press, Roe, G. and M. Baker 2007. Why is climate sensitivity so unpredictable? Science 318(5850): 629. Roseta-Palma, C. and A. Xepapadeas (2004): Robust Control in Water Management. Journal of Risk and Uncertainty, 29:1, 21-34. Starmer C. 2000. Development in non-expected utility theory: the hunt for a descriptive theory of choice under risk, Journal of Economic Literature XXXVIII: 332-382. Sunstein, C. R. (2005).Laws of Fear: Beyond the Precautionary Principle. Cambridge, UK: Cambridge University Press. Ulph, A. and D. Ulph 1997. Global Warming, Irreversibility and Learning, The Economic Journal 107, 636-650. UNESCO (2005): Precautionary Principle: World Commission on the Ethics of Scientific Knowledge and Technology, printed by United Nations Educational Scientific and Cultural Organization, France. Vogel, D. 2002.The politics of risk regulation in Europe and the United States.Yearbook of European Environmental Law 3: 1-42. Walker, B., S. Barrett, S. Polasky, V. Galaz, C. Folke, G. Engstrm, F. Ackerman, K. Arrow, S. Carpenter, K. Chopra, G. Daily, P. Ehrlich, T. Hughes, N. Kautsky, S. Levin, K.

24

THEPRECAUTIONARYPRINCIPLEANDGLOBALENVIRONMENTALCHANGE

Mler, J. Shogren, J. Vincent,T. Xepapadeas, A. de Zeeuw 2009. Looming Global-Scale Failures and Missing Institutions, Science, 325:5946, 1345 1346. Woodward, R., and R. Bishop (1997): How to Decide When Experts Disagree: Uncertainty-Based Choice Rules in Environmental Policy, Land Economics 73(4): 492507.