First of I am not a financial engineering guru. I worked as a COBOL programmer for many years and although I am a bit rusty with c++ I can do alright after a bit of a refresher.
I have have set myself the task of implementing ICA on a single stream of time series data (stock prices).
The ICA algorithm comes as part of the it++ package (fastICA) however it is necessary to perform some preprocessing on the time series data as ica works on the premise that the number of observations is at least as large as the number of sources.
From what I have discovered the preprocessing involves constructing a matrix Y that contains the original time series in the 1st column, a lag-1 shifted version of the time series in the 2nd column etc.
Y= 1.0135518 - 0.7113242 - 0.3906069 1.565203
- 0.7113242 - 0.3906069 1.565203 0.0439317
- 0.3906069 1.565203 0.0439317 - 1.1656093
My questions are:
What steps are necessary to recover the independent sources in the time series data after ICA is performed?
How to order the IC's from most significant to least significant.
If somebody could either answer these questions or point me in the direction of a suitable book or article I would be most thankfull. Preferably the book or article is more practicle in nature.
Log in to post a comment.