Research On Telecom Demand Forecast Model Based On Functional Data Analysis

The rapid development of telecom technology and the fierce market competition drive operators to increasingly rely on the demand forecasts for diverse of new telecom products to support strategic investment decisions. In academic, though demand forecasting model of traditional telecom products' are mature, the research of demand forecasting models for new telecom products is still in the exploratory stage (Robert Fildes, V. Kumar,2002). Diffusion models are widely used in the studies of new telecom products' adoption process at present; however, because of the lack and the sparsity of new services'data, the predictive performance of diffusion models need to be further verified. Ashish Sood, et. al (2009) first proposed an application of FDA (Functional Data Analysis) theory to discuss the penetration of new products. Although FDA has become a very important emerging field in statistics (J. O. Ramsay, B.W. Silverman,2002; etc.), it is not well known in the marketing literature. This study aims to first apply the cutting-edge FDA method to the field of telecom demand forecasting.Referring to the study of Ana M. Aguilera et al. (1997) and the Aili Qu (2009), we proposed a telecom demand forecasting model based on functional principal component analysis method based on the discussion of features of telecom demand data. The basic assumption of the model is that demand data of telecom products in different regions can be analyzed as functions, so the telecom products' demand curves can be represented by a linear model of mean function and the number of principal components functions. In addition, considering the differences of different telecom products' demand in different regions, this paper first took the cluster step and then set up the forecasting model by category during the modeling process.Three sets of new telecom products were chosen, and both of their time-series and cross-sectional penetration data were collected for the establishment of the FDA forecasting model, and the FDA model's predictive performance indicated a competitive edge of new telecom product forecasts by comparing model results with Time-series models, Bass model and its extensions.Meanwhile, a summary description was made to demonstrate the features of different classes of telecom products'adoption processes. First, the diffusion path of the new telecom services in the same area were saddle-shaped; second, the same telecom product launched in different regions can be divided into several classes demand forecast; third, there is a "lead-lag" influence of different telecom products launched in different areas. These conclusions will be a valuable reference to the prediction of new launched telecom products' demand in practice.