Accueil > Rubrique de services > Archive Equipe > Minh Ha Duong > Research statement to improve uncertainty analysis in integrated assessment (...)

Research statement to improve uncertainty analysis in integrated assessment modeling.

lundi 2 décembre 2002, par Minh Ha Duong

This research statement first summarizes past work before turning to expected contributions in the next two-three years and then to a longer research project with imprecise probabilities.

Past research

Theoretical research focused on environmental issues :
First, the core of my research is a reflection about uncertainty, precaution and sustainability in environmental decision-making and management. To this end I developed and maintain two integrated assessment models, each a few thousand lines of code : DIAM and IECCC.

- DIAM (Dynamics of Inertia and Adaptability Model) draws on macroeconomics optimal growth models such as Alan Manne’s Ramsey model (1986) or Bill Nordhaus’ DICE model. It is a stochastic decision with recourse model to analyze the timing of optimal carbon dioxide emissions reduction. It has been used to assess quasi-option values for various climate policy alternatives. Recently, with CIRED graduate student Patrice Dumas we examined a better probabilistic representation of catastrophic climate events using a kinked damage function in cost-benefit analysis. We showed that the possibility of bifurcations in climate dynamics in less than fifty years leads to more precautionary action in the short-term.
- IECCC (Integrated Emissions Composition Concentration Cost) is an stratospheric ozone policy model. With CNRS director Gérard Mégie and CNRS aeronomy researcher Dr. Didier Hauglustaine, we used it to publish of a counter-factual pro-active ozone layer protection scenario. We showed that acting ten years sooner would have allowed to prevent the formation of the antarctic ozone hole, at a moderate or negligible additional cost. We concluded that the Montreal Protocol was only partially successful at precaution : global atmospheric environmental problems could be regulated before surprising non-linearities occur.

Research on integrated assessment :

These models raise specific technical questions which are the second axis of my research activity. How to integrate large quantities of data coming from different sciences ? How to improve the quality of integrated assessment, especially for models embodying large quantities of technological or geophysical data ? While answers to these questions are epistemological, they cannot be reached without a hands-on approach, a constant practice of building models.

My research combines the action of building new models with longer-term reflections on methodology in the field. For example I assisted G. Begni at MEDIAS-France on writing an article about the remote sensed data needs for the Kyoto Protocol, and I published a paper on transparency and control in integrated assessment. The key aspect considered was how to reproduce models, studying both license agreements and practical issues. This paper was intended to improve the place of integrated assessment within the scientific community. This is all the more important it is a young and applied research discipline.

I am striving to improve my integrated assessment process by borrowing both from experimental sciences and from software engineering. I take advantage from the fact that scientific manuscripts are information products too, just like code. This reflection already led to practical improvements, such as using revision control tools such as RCS and using project management tools such as make.

Outreach activities :

The third axis of my work is to carry the policy-relevant implications of integrated assessment models to the public arena. Disseminating results at the non-technical level is not part of a research program if it is seen as a pure intellectual quest. But from a more general point of view it is an integral part of researchers social duties. Expertise with environmental issues led me to write essays on policy-relevant topics, non-technical papers targeting a broader readership. There model results are used to illustrate the ideas justified by rigorous but informal reasoning.

In this line, I disseminated my doctoral work focused on climate change and cost-benefit analysis by publishing in French and Vietnamese peer-reviewed interdisciplinary journals. A jointly authored paper summarizing the results of the INASUD research project on climate change policy has been published. And with Jean-Charles Hourcade, we wrote a survey discussing the issues of controversy and target-setting, timing, intergenerational equity, negotiations and policy instruments, to appear as the topic paper on Policy Framework for Climate Change Integrated Assessment for the Encyclopedia of Life Support Systems.

Contributions in the pipeline

Currently under review :

Two submitted manuscripts are the outcome of a collaboration with David Keith (Carnegie Mellon University) started during my research visit. In one, we show the possibility of capturing carbon dioxide from the air at a price lower than 200 dollars per ton of carbon, and explore the (mostly long-term) implications for climate policy. The other paper is about sequestration of carbon dioxide by injection into geologic formations.

Sequestration raises a sustainability issue because some carbon is likely to leak back into the atmosphere. Therefore, in a normative economics analysis the discount rate plays the central role. My work makes explicit the trade off between discounting, leakage, the cost of sequestration and the energy penalty necessary to capture, transport and inject carbon underground. Results show that the value of a certificate for `Temporary long-term underground sequestration’ should be less than the value of a certificate of `Carbon emissions reductions’, with a ratio at about eighty percent for a leak rate of 1% and a discount rate of 4% per year. The energy penalty of carbon capture and sequestration should be deduced from this ratio.

Third, I recently resubmitted a paper titled `Introduction to economic approaches of uncertainty’ in French. It targets an interdisciplinary readership interested in learning how economists study uncertainty, climate policy and integrated assessment models.

Ongoing collaborations :

Turning now to work in progress, Dr. Benoit Morel (Carnegie Mellon University) and I have studied the real option with a down and out barrier. It models an investment decision with a double irreversibility concern : investing is irreversible, but waiting runs the risk of loosing the opportunity to invest. This applies to dot-coms for example, which can be de-listed from the stock market when the share value gets below one. We demonstrated that the optimal strategy leads to earlier investment when the barrier increases, or when uncertainty decreases. Uncertainty has ambiguous effects on the expected decision time and on the investment probability after N years.

With respect to equity issues, Dr. Nicolas Treich (INRA, France) and I are working on risk aversion and time preferences. Using intergenerational recursive preferences, we examine both theoretically and numerically how social preferences with respect to equity and risk affect the optimal management of a stock externality. We show that higher relative risk aversion increases short run control of pollution. On the contrary, resistance to intertemporal substitution and discounting tend to decrease today’s pollution control. A climate policy model is used to explore the sensitivity of an optimal energy tax to the utility function. This sensitivity is several times more important when equity and risk preferences are separated than when they are assumed equal.

Finally I recently started a research with Dr. Elizabeth Casman (Carnegie Mellon University) on a project led by Prof. Granger Morgan (Chairman, Engineering and Public Policy Department, Carnegie Mellon University). The goal is to use available toxicological epidemiological evidence, together with expert elicitation, to bound the number of lung-cancer fatalities attributable to one or several suspect agents. The method is to generate these bounds as a residual after accounting for better-known causes such as smoking, radon and asbestos, using a principle of maximum uncertainty. This project is an opportunity to apply imprecise probabilities, a theory that I intend to develop more in the future.

A research project on imprecise probabilities

Early in my research career I decided to start by studying the classical probabilistic framework : stochastic dynamic programming and real options, before turning to alternative paradigms such as fuzzy numbers, the theory of evidence or the Choquet integral. At that time ten years ago it was unclear for me which theory would be a promising alternative to the standard Subjective Expected Utility model.

Now in my eyes imprecise probability theory is the new paradigm that generalizes and unifies all the other theories mentioned above.

Imprecise probabilities can be defined by giving a convex set C of probability distributions. This allows the following definition of precautionary decision-making, that is the basis of my future research : gain a is preferred to b if and only if the expected value of a is larger than the expected value of b for all probability distributions in C.

Different shapes of C correspond to different theories of uncertainty. Classical risk analysis is the special case where the set is just one distribution. In this way the turn to imprecise probabilities is not a rupture from, but an extension of previous decision-making models.

Prospective scenarios are often used to analyze long-term questions where uncertainty is a major issue such as climatic change. The problem is that most often scenarios are presented symmetrically, without quantifying any differentiated degree of confidence, despite the fact that some combinations of values, such as high per capita income growth and high population growth, are less likely than other combinations. There is a gap between scenarios, where likelihood is unquantified, and forecasts, which have probabilities. Imprecise Probabilities is proposed to bridge that gap, as a philosophico-mathematical theory of uncertainty appropriate for long-term prospective scenario-based analysis.

I propose to elaborate a theory of strategic planning, where the likelihood is quantified with imprecise probabilities, by formalizing the following principles : 1. Maximize the scenario set plausibility ; 2. Minimize the plausibility that the scenario set does not bracket the future ; 3. No scenario should be preferred to another according to the precautionary decision-making criteria ; 4. Distort information either by generalization or by omission, but not both. 5. Minimally distort information and 6. Maximize the information content.

A pilot study has been conducted at a reduced scale. I used empirical data from the IPCC integrated assessment models database and the Keith-Morgan expert elicitation study. This allowed me to formally attribute a degree of surprise to each scenario and quantify to which extent the Intergovernmental Panel on Climate Change (IPCC) scenarios sets represent the underlying uncertainty.