On the assessment of economic risk: factorial design versus Monte Carlo methods

Abstract

The feasibility of large investment projects (such as gas transmission and power system projects) has many aspects. Usually, this problem cannot be modeled as a single optimization problem; instead, the multiple aspects (demand, supply, prices, investment costs) are modeled separately. Each aspect may require a large, nonlinear submodel. The results of such a submodel can often be summarized by one or a few variables, which combine all the submodel's information; for example, total demand is the sum of the demand per customer type, each type being modeled separately. Traditionally, the feasibility of the investment project is then judged by combining the results of the various submodels for the ‘base case’ values of all model inputs. This base case information, however, is not sufficient for the decision makers; they also like to know the economic risk they are taking. To assess this risk on the project level (Hertz, D. B., Risk analysis in capital investment. Harvard Business Review, 1964, 95–106) developed a method known as risk analysis. This method is based on the estimated probability distribution of a project's net present value ( NPV). This distribution is obtained by introducing distributions for the model inputs. The project's economic risk is then. expressed as the probability of a negative NPV exceeding a critical value (say) α. Nowadays this approach is becoming popular, because many software packages (such as @RISK and Crystal Ball) facilitate such a risk analysis. Although Hertz's risk analysis is appealing, it has a number of theoretical and practical flaws, which may lead to wrong conclusions. These flaws are discussed in this paper. From a modelling point of view, Hertz's risk analysis is similar to analysing the technological or operational risk of an investment. However, economic risk and technological risk are different concepts that require different analyses. In this paper these differences are discussed and it is shown that Hertz's risk analysis does not measure what is normally meant by a project's economic risk. Furthermore, the information requirements for the application of risk analysis to large investment projects are formidable; this makes the results of Hertz's investment analysis unreliable. Less information is required by sensitivity analysis based on the statistical design of experiments (such as 2 k− P designs); this analysis is more robust, and leads to results that better satisfy the information needs of decision makers.