Toward an Appropriate Measure of Uncertainty in a Risk Programming Model

Abstract

Abstract The measure of uncertainty in a risk‐programming problem has long posed a dilemma. The use of the variance of realized returns assumes that the distribution of realized returns is the same as the distribution anticipated by the decision maker prior to the start of production. Rejection of this maintained hypothesis requires either direct elicitation of these distributions or construction of another measure from realized data. A mean‐squared forecast error is considered an appropriate measure. Optimal solutions to a quadratic risk‐programming problem are obtained using this measure and compared to those obtained using traditional measures.