TESTING THE ALGORYTHM OF THE “CATERPILLAR”-SSA METHOD FOR TIME SERIES RECOVERY

Abstract

The basic algorithm of the “Caterpillar”-SSA method is considered and tested.

Full Text

One of the significant problems in the analysis of time series is the separation of trend and periodicals presses from the noise. This research is about a robust method of time series analysis: “Caterpillar”-SSA, which is currently being developed. Let’s investigate the functioning of this algorithm and state, in what its specificity is exactly. The variant of the algorithm described below doesn’t essentially differ from the basic one [1], it has only been simplified without any changes in result. We consider the given time series F: 0 1 1 , ,..., N f f f − , (1) where N is its length. Further we assume that F is a nonzero series. The algorithm consists of four consistent steps: investment, singular decomposition, grouping, and diagonal averaging. The investment procedure converts the time series F into a sequence of multidimensional vectors called the trajectory matrix. Mathematics, mechanics, computer science 140 To analyze the time series we select parameter L called “the length of period”, which is in the open interval 1 < L < N. Thus K = N−L−1 investment vectors are created: 1 2 ( , ,..., )T , 1 i i i iL X f f f i K − - − = ≤ ≤ . (2)