Comparison of singular spectrum analysis and singular spectrum decomposition based on new trajectory matrix in reconstruction and forecasting the time series.

Singular Spectrum Analysis (SSA) is a new powerful method in time series analysis. This non-parametric method due to its unique properties, such as there being nonecessity as to making assumptions about stationarity of time series and also about the normality of residuals, has caught the attention of many researchers in the field of Econometrics and time series analysis and its applications are increasingly getting wide spread. Also this method can be used for short time series. The main purpose of SSA method is to decompose time series into interpretable components such as trend, oscillatory component, and unstructured noise. The basis of SSA is singular value decomposition of the trajectory matrix built on the time series. In the basic SSA method the frequency of observations which used in the trajectory matrix is different and so there may be an error in reconstructing and forecasting the time series, especially at the beginning and end of the series. It occurs because the magnitude of eigenvalues, eigenvectors, and consequently, reconstruction and forecasting of future values of time series, is directly related to the trajectory matrix. The purpose of this paper is to improve the trajectory matrix of SSA method to increase the accuracy of the reconstructed time series and forecasting results, which is called singular spectrum decomposition (SSD). In this paper, SSA and SSD methods and their properties are briefly introduced and then the performance of SSD method over SSA method in time series reconstruction and forecasting for simulated and real data is discussed.