Box-Jenkins vs Multiple Regression: Some Adventures in Forecasting the Demand for Blood Tests

This paper reports the application of a multiple regression forecasting model in a hospital laboratory setting. Twenty-five time series models were tested, and the comparative results should be of interest to practitioners. Box-Jenkins theory could not be stretched to fit the time series, which is characterized by powerful but erratic trend and seasonal components. Simple exponential smoothing performed much better than several highly sophisticated smoothing models.