Some Observations on Adaptive Forecasting

The procedure of “adaptive” or “exponential” forecasting is based on weighted averages of two sources of evidence; one is the latest evidence the most recent observation, the other the value computed one period before. As such, it is an easy, quick and cheap method; very little information is needed for a forecast; also, the most recent information is used. This article serves a dual purpose. One is to simplify the forecasting procedure and to clarify its characteristics in the simplest possible manner. This objective is pursued in Sections 2--4. The second purpose is to formulate a probabilistic model underlying the prediction procedure and to select weights which minimize the mean-square prediction error. Sections 5 and 6 are devoted to that purpose.