Model Selection in Demand Estimation in a Forecasting Environment

• Activity Number: 84
• Type: Contributed
• Date/Time: Sunday, July 31, 2011 : 4:00 PM to 5:50 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Burcu Aydin*+ and Kemal Guler and Enis Kayis and Mehmet Sayal
• Companies: Hewlett Packard Laboratories and Hewlett Packard Laboratories and Hewlett Packard Laboratories and Hewlett Packard Laboratories
• Keywords: Model Selection ; Switching Model ; Data Augmentation Algorithm ; Collaborative Inventory Management

Abstract

The accuracy in forecasting is a crucial success element in Collaborative Inventory Management(CIM) and other business practices that employ forecasting. The management of inventory levels based on forecasts issued by buyer and suppliers is a core CIM concept. The task of measuring the accuracy of these forecasts can get complicated when some of the necessary demand information is missing. This study handles the problem of model selection in a Switching Model devised to estimate unobserved demand in a CIM setting. The proposed solution of this model involves an adaptation of the Data Augmentation Algorithm(DAA), where the imputation step requires the application of multiple linear regression method to a submodel. The extensive amount of data that can be used as explanatory variables in this submodel necessitates embedding of suitable model selection methods within the iterations of DAA. Our focus in this work is to explore various model selection methods within the DAA framework. We provide numerical analysis results on effects of each method on a real-life data set. Also, we aim to provide theoretical results on effect of using these methods on the convergence of DAA algorithm.