Horizontal Forecasts

Abstract

Perhaps the most typical demand pattern is the horizontal where the month-to-month demands fluctuate above and below a path (called the level) without any trend or seasonal influence. This chapter describes five horizontal forecasting models. These forecast models are here called the following: horizontal forecast, horizontal moving average forecast, horizontal discount forecast, horizontal smoothing forecast, and forecasts using 2 stages. In all situations, the concept of raw and integer forecasts is shown. For each of the models, monthly raw forecasts are generated in fractional form. A corresponding set of forecasts is called integer forecasts and these are converted from the raw forecasts by way of the rounding algorithm. A key measure of the forecasts is the standard deviation of the 1-month forecast errors. This measure is needed subsequently when inventory decision are computed. Another useful measure, the coefficient-of-variation, is a relative way to measure the forecast error.

Summary

Five horizontal forecast model s are described. The horizontal forecast model is based on the N most prior monthly demands where each demand entry is assigned the same weight in generating the forecasts . The horizontal moving average forecast model uses a parameter N that specifies the number of most recent monthly demands to use in generating the forecasts. The method gives equal weight to each of the demands. A special adjustment is made for new parts when the number of history months is less than the parameter of N. The horizontal discounting forecast model is based on two parameters, N, the number of months of history to use, and β, the discount rate that assigns more weight to each more recent demand. The horizontal smoothing forecast model also gives more weight to the more recent demand entries. The model adjusts the prior forecast with the most current demand entry. The horizontal 2-stage forecast model generates a forecast for the number of lines by future months and then applies the average pieces-per-line to obtain the forecast of demands for each of the future months. This method may be useful when the history demands are of the lumpy type. For each of the forecast models, the methods generate raw forecast s that are in fractional form. A cumulative round algorithm converts the raw forecasts to integer forecasts . The standard deviation of the 1-month forecast error is also computed for the models. This is a measure on the accuracy of the forecasts and is needed in subsequent computations concerning the inventory replenishments in the stocking location. A relative way to measure the forecast error is by the coefficient of variation , denoted as cov.