On the Bayesian Steady Forecasting Model

Summary A recent criterion of Smith (1979) seeks to generalize the steady forecasting model of Harrison and Stevens (1976) to take account of non-normality in the observation and system variables. In this note, predictive distributions and forecasts corresponding to quadratic and step loss functions are examined when the observation variable has a distribution belonging to the exponential family. It is shown that Smith's generalized model does not always yield forecasts which satisfy constraints obtained from the more familiar forms of steady forecasting. Two particular cases of the exponential family are used to illustrate the point.