Forecasting the failure of component parts in computer systems: A case study

This paper is an applied study in forecasting the failure of component parts in computer systems to aid in production planning and inventory control. The aim is to develop a reasonably simple forecasting system that can be operated by managers rather than statisticians. Monthly failures of components are shown to be related to cumulative shipments from the factory after a time lag of several months. This relationship is modelled using discounted-least-squares regression, a methodology that appears to be rare in practice. Simulated forecast accuracy is then compared with two alternatives, exponential smoothing and a combination of regression and smoothing forecasts. Discounted regression proves to be the most accurate and is therefore implemented.