Economic Index Forecasting via Multi-scale Recursive Dynamic Factor Analysis

Abstract

In this paper, we propose a new multi-scale recursive dynamic factor analysis (MS-RDFA) algorithm for economic index foresting (EIF). The proposed MS-RDFA algorithm first employ empirical mode decomposition (EMD), which is a powerful tool for multi-scale analysis and modeling on the non-linear and non-stationary signal such as economic index data. Moreover, an efficient RDFA algorithm using recursive subspace tracking is adopted to explore the correlated nature of the adjacent intervals of the economic index data. The one-step prediction of PC scores is modeled as an AR process and can be recursively tracked by Kalman filter (KF). The major advantage of the proposed MS-RDFA method is its low arithmetic complexity and simple real-time updating, which is different from other conventional algorithms. This makes it as an attractive alternative to other conventional approaches to EIF on mobile services. The experiments show that the proposed MS-RDFA algorithm has better forecasting results than other EIF methods.