MUSTARD: Multicriteria utility-based stochastic aid for ranking decisions

✍ Scribed by Yannis Siskos

• Publisher: John Wiley and Sons
• Year: 2002
• Tongue: English
• Weight: 159 KB
• Volume: 15
• Category: Article
• ISSN: 0894-3257
• DOI: 10.1002/bdm.422

✦ Synopsis

The outcomes of investment projects are generally characterized by a certain amount of uncertainty since they depend upon the future global economic circumstances, the potential users' reactions, the actual technological performance, as well as many other factors that are not well known. To that extent investment projects involve some risk which deserves attention at the decision-making stage. This is also the case for public investment projects, despite the fact that the state has a large investment portfolio, which induces some risk spreading. In effect, their risks often cannot be averaged out because they bear particularly upon some subset of population or equally upon all people. Nevertheless, in actual practice of public investment assessment, the handling of risk is most often rather cursory and limited to sensitivity analysis exercises. However useful that technique can be, it does not solve the decision makers' problem since it only simulates the outcomes of possible scenarios or alternative assumptions, and does not help decide whether the risk is worth taking. Beside the decision makers' reluctance to reveal their own preferences and attitude towards risk, one important problem is the difficulty of estimating a non-linear preference function in a context of uncertainty, a prerequisite for assessing attitudes towards risk. This difficulty is particularly great when the decision problem is set in a multicriteria analysis framework. In that case, some available techniques, like MAUT (Keeney & Raiffa, 1976), do not appear practical enough for actual use in public project assessment. This is because eliciting a decision maker's preferences tends to lead to a difficult and lengthy questioning process, and it can become very arduous indeed when preferences between uncertain projects or lotteries must be decided.

The software package MUSTARD embodies several models and techniques, which aim at solving this difficult problem for an additive utility function. First, in order to reduce the number of parameters to be estimated, the specifications of the additive utility assessment methods UTA or the Quasi-UTA are used, where the non-linear partial 'utility' functions are made of a small set of linear segments. A maximal reduction of the number of parameters is obtained with the Quasi-UTA specification where the functions are structured as recursive exponential functions of only one curvature parameter. Note that the UTA methods aim at inferring one or more additive value functions from a given ranking on a reference set of projects. The methods use special