Assessing risky public investments with MUSTARD

This paper presents a practical implementation of multicriteria methodologies based on the utility additive (UTA) model by Jacquet-Lagrèze and Siskos and Quasi-UTA model by Beuthe and Scannella, which are specified with a non-linear, but piecewise linear, additive utility function. In contrast with the general UTA model, the Quasi-UTA specification structures the partial utilities as recursive exponential functions of only one curvature parameter. This allows for a reduction of the quantity of information necessary to build the utility function. The software MUSTARD implements different variants of these models. Firstly, it offers the basic deterministic UTA model of disaggregation, but also its first programmed stochastic version. In both cases, the software proceeds stepwise and interactively helping the decision maker to formulate the problem and state preferences between projects; in the stochastic case, the decision maker is even helped to build the criteria distributions. The Quasi-UTA specification can be introduced in this disaggregation model. Secondly, the software offers an aggregation model whereby the Quasi-UTA partial utility functions are built separately through specific questioning processes. The questions relating to deterministic criteria are of the ‘direct rating type’, while those of the stochastic criteria are either of the ‘variable probability’ or the ‘variable outcome’ type. The criteria weights can be assessed by the ‘swing weight’ method or by a UTA-II side-program. As an example as well as a test of the Quasi-UTA aggregation approach, the paper presents its application to a real problem of selecting road investment projects in Belgium. Several experts and civil servants were interviewed, and their individual utility functions derived. The projects are ranked according to their rate of return, which is computed on the basis of the projects' certain equivalent money value.