The issue of forming a composition of estimators, which can be forecast from quantitative predictive models or expert opinions, is discussed and a case made for linear combinations. Two methods are presented, one aimed at an optimal, minimum variance composition; and the other, at utilizing linear weights with a directly meaningful probabilistic interpretation. Bayesian estimation methods are used in both cases. © 1979.
Mendeley saves you time finding and organizing research
Choose a citation style from the tabs below