FORECASTING ERRORS USING MAD

The paper is a study of the effects of using mean absolute deviation MAD to estimate variability in setting reorder levels for the inventory of a stock item. The method presently employed by NavSup in setting such reorder levels involves exponentially smoothed estimates of the mean and variance of the demand process. Any error involved in setting reorder levels results in a change in the underlying risk which in turn can be translated into costs. Such errors for the method of estimation presently employed are compared with standard maximum likelihood procedures. By simulating several normal systems, the smoothing technique is found to be inferior to classical methods with no reduction in computational difficulties. Author