The Adaptability Of Panel Data Models For Telecom Demands: An Empirical Study

With the gradual opening up of mobile Internet and the trends of three networks convergence, new types of data services show rapid growth and increase the complexity of telecommunications demand which make the prediction efficiency of different prediction methods a challenging research topic. Panel data model is an important development of econometric theoretical methods in recent years which is characterized by combining the advantages of time series data and cross-sectional data. Also the model of panel data is more diverse and with more features. This paper has demonstrated that certain types of service prediction need to adopt certain type of panel data to improve the prediction accuracy, further to investigate the adaptability of different telecommunications demand models.This paper has accomplished the forward prediction of fixed telephone users, mobile phone users, Internet users, mobile phone charging duration, the amount of mobile messaging, the amount of weather service, the amount of MMS billing and mobile data traffic using Panel data model, time series forecasting models, cross-sectional data regression models and innovation diffusion model which based on factors affecting the telecom demands and the sample data of telecommunications business volume and the number of users.Firstly, the establishment of telecommunications demand forecasting model of panel data. The author of this paper established Telecommunications requirements panel data model including mixed model, the static fixed effects model, static random effects model and dynamic panel data model based on summarizing the macroeconomic factors of telecommunications needs such as GDP, the level of urbanization, income and consumption levels and so on, and based on the consumption function of Keynesian’s, and then described the different models of estimation and testing methods.Secondly, an empirical study on different services based on the telecommunications needs of different panel data models. In order to study the adaptability of different telecom services on the panel models, This paper selects the fixed telephone users, mobile phone users, Internet users, mobile phone charging duration and the amount of mobile messaging of the entire network, and the amount of weather service, the amount of MMS billing, mobile data traffic of a carrier as the dependent variable, and then estimated and tested Hybrid model, the static fixed effects model, static random effects model and dynamic panel data model. Based on the parameter estimation results, this paper has accomplished the forward prediction for different telecom services, and compared the prediction accuracy of different models.In addition, verify the superiority of the panel data model. In order to test the superiority of panel data model, this paper established the ordinary time series model, cross-sectional data regression model and innovation diffusion model for telecom services. At the same time based on the parameter estimates results, this paper calculated the prediction accuracy of different models. The Study found that the prediction accuracy of the panel data models’were higher than ordinary time series model, cross-sectional data regression model and innovation diffusion model.Finally, the study demonstrated that, because of considering the cross-section data and time series data and the strict control of individual heterogeneity, the panel data models are better than the ordinary time-series forecasting model or cross-sectional data regression model in forecasting. In the static panel data model, the fixed effects model is better than random effects models and hybrid model in prediction. In comparison with fixed effects model and the dynamic panel data model, it can be found that the fixed effects model are suitable for traditional telecommunications demand forecasting, such as the number of fixed telephone users, mobile phone users, the number of Internet users, mobile phone charging duration and the amount of mobile messaging;and the dynamic panel data model are suitable for new telecommunications services or low penetration of telecommunications services forecasting.