A Method for Fast and Robust Sungular Spectrum Analysis

One of the main problems encountered in the analysis of real-world time series, is the existence of outliers originating from various sources other than the dynamics of the series. Such sources can be a momentary noise or an imperfection in the recording devices. This problem has led to the need to construct robust methods using metrics other than the usual mean squared error. In combination with non-parametric techniques such as the very popular Singular Value Decomposition (SSA), it can give iterative algorithms that deal with this problem but create huge needs in computing power, especially in problems where we need fast response of the system (for example in high frequency trading). In this paper, a technique addressing both problems is presented which exploits the peculiarity of the Singular Value Decomposition (SVD) analysis of square symmetric matrices.