Demand & Inventory Management Essay

Words: 5020
Pages: 21

Forecasting demand and inventory management using Bayesian time series

T.A. Spedding University of Greenwich, Chatham Maritime, Kent, UK K.K. Chan Nanyang Technological University, Singapore

Batch production, Demand, Forecasting, Inventory management, Bayesian statistics, Time series

Keywords

Introduction
A typical scenario in a manufacturing company in Singapore is one in which all the strategic decisions, including forecasting of future demand, are provided by an overseas office. The forecast model provided by the overseas office is often inaccurate because the forecasting is performed before the actual production schedule and it is based on marketing survey results and historical data from an overseas research team. This
…show more content…
To illustrate the demand forecasting techniques, a case study of a manufacturing company is presented in the paper. Three different forecasting models with three different sets of data (the first 24 weeks, the first 36 weeks and the complete 51 weeks data) are built using Bayesian dynamics time series and forecasting and ARIMA techniques. The three different sets of data are used to investigate the dynamic behaviour and structural change of the system. From the modelling results, a comparative study of the models' accuracy is performed. The strengths and weaknesses of the Bayesian time series analysis (BATS) models are also discussed and compared to the ARIMA model.

The Bayesian dynamic linear time series model
Traditional forecasting approaches are based on characterising the structure of historical time series and then predicts future events based on that structure. Obviously, the structure of the time series may change in a volatile business environment. The parameters of the time series model would then need to be re-estimated based on the new structure of the time series. Bayesian Forecasting, however, is based on the principle that routine forecasts can be updated by subjective intervention as external information becomes available. In this context ``Bayesian learning'' is particularly relevant to applications in a dynamically changing environment. Bayesian dynamic time