A DEMAND FORECAST METHOD FOR THE FINAL ORDERING PROBLEM OF SERVICE PARTS

摘要

Demand forecast of service parts at the end-of-life phase of durable goods is plagued with inadequate demand data, changing purchasing behavior, and lack of reliability information. As the number of sale data for each part is very limited, conventional forecast methods are not applicable. This paper presents an empirical study on developing a forecast method based on installed base information. Several archetypes of demand trend are first identified and regular regression is shown to be inadequate in predicting future demand. Then by applying the installed base approach, the interrelated effects of data trend, data quantity and data recency are unraveled. This knowledge enables a new forecast method to be developed based on two tests of data trend. It is found that for parts with an upward trend it is better to use more data and apply linear regression but for parts without a trend it is better to use less but more recent data with a constant regression function. The proposed method is validated with multiple automobile and notebook computer series and is shown to outperform a current method by large margins in forecast errors.