Demand Forecasting of Automobile Spare Parts after the End-of-Production - A review of demand forecasting models

Abstract: Demand forecasting of spare parts plays a crucial role in automobile industry where it generally requires a significant attention in controlling inventory. It is possible to maintain an optimal stock level when there is a continues supply at the Original Equipment Manufacturers (OEMs). However, parts supply is expected to be interrupted after the end-of-production (EOP) of a vehicle model, and this study will focus on forecasting spare parts for this period. After an EOP of a vehicle brand, the company requires to supply spare parts for a long period of time to support its aftermarket, therefore a long-term forecast for spare parts to support this period is needed. The researchers observed that there is a significant knowledge-gap in this area and conducted this research with collaboration of a new automobile brand in Sweden. The purpose of this research is to investigate the potential demand forecasting methods that already exist and select the most accurate and suitable model that would answer the research question of selecting the best performing demand forecasting model for EOP phase. The research aims to list the demand forecasting models, test and rank the models, and provide recommendations based on the findings though a qualitative research approach which employs a systematic literature review according to PRISMA method. The models were tested using actual spare parts demand data of the company and have ranked the models according to the Pairwise Comparison in Analytic Hierarchy Process (AHP). The models are then ranked based on five criteria that are supposed to answer the research question. They are namely, accuracy, data requirement, time horizon, user-friendliness, and generalizability. The results shows that Exponentially Moving Average (EWMA) is the best performing model out of the ten model with an overall AHP score of 16%, followed by Theta (13%), Holt's Winter (12%), and Single Exponential Smoothing (11%). It was also noticeable that some of the popular models such as Croston's and ARIMA models ranked below the benchmark (Naïve forecast). As per the error tests of the four demand types, EMWA still performs better than the other models. Thereafter, Holt-Winters and ARHOW models perform better for erratic, intermittent and smooth demand types while Theta and SES are most suitable for the lumpy demand. The findings of this research are generalizable to the other industries that use spare parts and important to the academia as these open many future research opportunities.