Statistical correction of judgmental point forecasts and decisions

In many organizations point estimates labelled as ‘forecasts’ are produced by human judgment rather than statistical methods. However, when these estimates are subject to asymmetric loss they are, in fact, decisions because they involve the selection of a value of with the objective of minimizing loss. While there are often considerable advantages in using judgment to arrive at these decisions the psychological demands of the task may mean that the resulting decisions are sub-optimal when compared with those resulting from a normative decision model. In these circumstances a combination of statistical methods and judgment may be superior. This paper suggests a procedure which involves the statistical correction of the original decision to obtain a forecast and the subsequent use of a mathematical model to identify the theoretically optimal decision in the light of this forecast. The application of the procedure to the monthly decisions of a manufacturing company suggests that it may offer the potential for achieving substantial improvements in many practical contexts.