Forecasting of core inflation

Linyun Zhang, Jinchang Li and

Abstract

Nonparametric support vector regression (SVR) approach has recently attracted much attention as a new technique for forecasting in economics. The chief advantage of this new approach is that such models are free from the large sample assumption that is often adopted to make the traditional models. In addition, SVR inherits the strong fitting ability of neural network and its minimum structural risk endows it stronger predictive ability than neural network. This paper uses SVR approach to predict core inflation series estimated with variance paring, compares forecast result with that obtained by maximum likelihood estimation (MLE) and back propagation (BP) neural network. It suggests that SVR presents the best predictive ability. On this basis, this paper uses SVR approach conducts trend extrapolation on core inflation and concludes that China will continue with inflation in the next year but such inflation will fall slightly.