Independent Component Analysis vs Single Spectrum Analysis

[p75213]

Homework Statement

Hello,
I have implemented single spectrum analysis and independent component analysis against a time series (in this case the euro foreign exchange) as a smoothing indicator. The ssa algorithm I purchased from: Caterpillar-SSA - http://www.gistatgroup.com/cat/programs.html and the ica algorithm is fastICA from it++ open source software. In both cases the window size is set at 90. SSA has 6 principle components and fasICA has 6 independent components.

In both cases the time series is deconstructed into a matrix of delays before either the ssa or ica processing takes place. After processing the smoothed indicator is reconstructed using diagonal averaging. The strange thing is in both cases the indicators are nearly identical (not quite - observe at the very end where they diverge slightly). Should this be the case?
Also as new bars are added to the series both indicators readjust themselves retrospectively. How can this action be minimised or better still stopped in either indicator.

Homework Equations

The Attempt at a Solution

 

[Valkarie]

The fact that the SSA and ICA indicators are nearly identical is likely due to the fact that both algorithms are trying to identify similar patterns in the data. Since they are both based on matrix decomposition, they are looking at the same underlying structure in the data. In order to minimize the retrospective adjustment of the indicators as new bars are added to the series, you can try using a window size that is larger than 90. A larger window size will provide a more steady indicator and reduce the amount of adjustment that needs to be done. You could also try using different processing algorithms such as wavelets or Fourier transforms.