Home / Research / Cluster 1 / Theme 4

Quantitative models for policy and decision-makers

Bocconi University

Emanuele Borgonovo
Associate Professor
Department of Decision Sciences
Bocconi University
Viale Isonzo 25, 20135 Milan, Italy
emanuele.borgonovo@unibocconi.it



Bocconi University
At a regulatory, practitioner and academic levels, the use of decision support models that allow the understanding of the behaviour of economic, demographic and natural systems has become widespread. Model results are an integral part of the decision-making process. They aid policy and decision-makers in the identification of optimal policies and/or in predicting the evolutions of these systems given changes in current policies. However, notable examples (see also the recent article of Greenspan, 2008 published on Financial Times) show that an understatement of uncertainties or an overstatement in a models prediction capability can lead to severe consequences not only for the decision-makers, but also for the public.

In order to cope with the difficulties in model creation and model utilization, recent research is focusing on the following aspects:

Uncertainty Quantification
In the use of a quantitative model for policy or decision-making the determination of the analysts¿ degree of confidence in model results is crucial in avoiding that ¿an underestimation of predictive uncertainty [Saltelli et al, 2004]¿ results in overconfidence in model results. Uncertainties affect various levels of the modelling process and needs to be incorporated in the model or separately assessed, if left outside the model.

Model Validation
Modern models are complex (refer, for example, to the sophistication of a full-fledged professional financial model or of an agent-based model describing a chaotic system) and a plethora of numerical calculations interpose themselves between the model result (i.e., what the decision maker ¿sees¿ and ¿uses¿) and the exogenous variables that determine the model result. It is, therefore, most of the times impossible for model users to have a thorough understanding of the model¿s structure. Such lack of knowledge is then associated with modelling risk, i.e., the risk that ¿the output of the model is incorrect¿ (Fabozzi, 2000). Therefore, there is need for model validation and for the use of proper techniques to assess model reliability.

Factor Prioritization
In many relevant applications, there it is desirable to restrict attention and resources to a subset of factors (the key drivers). An analysis based on the sole expert judgment could lead to the a-priori exclusion of relevant factors. There is, then, an urge to make the selection systematic.

Policy Evolution Monitoring
Model results are often utilized to gain indications of optimal policies. Assumptions are monitored over time (usually with a given period). Changes in the optimal strategy can be realized as an effect of the new information. There is then the need of techniques capable of providing analysts with a methodological approach to understand and explain the difference in model results and the changes in optimal policies through time.

These problems need to be addressed by the use of appropriate modelling, sensitivity and robustness analysis methods capable of aiding the modeller in coping with model complexity. Research in this direction is ongoing in several disciplines (ranging from Operations Research, to Economics, Finance and the Management Sciences).

Relevance to future of European societies
As we mentioned in Section, the understanding of model behaviour and limits is essential in several decision and policy-making sectors. In His Financial Times article of March 16 2008 concerning the subprime crises, Alan Greenspan writes that ¿we will never have a perfect model of risk¿. The article underlines how a partial understanding model of a models predictions capability can lead to highly significant financial and social consequences. Greenspan¿s observations suggest that models results need to be integrated by a corresponding scientific approach allowing: 1) a full understanding and explanation of the model output and 2) an assessment of the model limitations, e.g., of the scenarios in which a model¿s results are reliable. Such goals, if reached, would have a direct relevance in all those Sectors where models are utilized to support policy and decision making. At an European Commission level, for instance, policy makers (decision making bodies and regulatory agencies) would be equipped with improved predictions capabilities, with a better understanding of the evolution of economic and financial systems, which will also allow an improvement in the decision-making process.

 
© 2009 - 2011 RISE Network