Edited by xD&Cox, 05 January 2007 - 06:00 PM.
Oil and Oil Services trade update...
#1
Posted 05 January 2007 - 05:57 PM
#2
Posted 06 January 2007 - 09:55 AM
This is the normalized volatility patterns according to the 95% confidence interval (3 standard deviations) during the logarithmic rise of the 2002 - July 2006 period.
The patterns are better extracted during a secular trend, here's the 75% of the pattern intensities during the rise. There is visible right translation in the cycles, I suspect it will change and shift to the left if the secular trend has turned or turning to neutral for now [LINK]. So the above chart might favor more of the waterfall patterns during the declines...
Notice that the trend has indeed changed since the bounce has been considerably weak in the second half of the 2006, but the volatility path favors the upside or stable prices during the first half of the 2007...
I posted the periodicity transform paper here a while back...
- kisa
Edited by kisacik, 06 January 2007 - 10:00 AM.
#3
Posted 06 January 2007 - 02:13 PM
#5
Posted 06 January 2007 - 08:31 PM
- kisa
Edited by kisacik, 06 January 2007 - 08:33 PM.
#6
Posted 07 January 2007 - 10:37 AM
I must tell you, unlike principal component analysis (KLT) or any other orthogonal method, PT requires an understanding of what is important in the data and its decomposition. There is no unique solution, I favor maximizing the signal to noise ratio by using my own metrics, but I do not decompose the entire data in one pass or large matrix. My methods are iterative and it will penalize and rebalance. But, I do decompose into the likely patterns that are not orthogonal too, but the real strength of my method comes from the ability of the evaluation of the likely momentum and the advection within the likely periodicities, or patterns, in multiple time frames. I actually do not assume any basis functions or patterns strictly either...
- kisa
So basically you take several PTs of different time frames using different parameters and you measure the likelihood of each by its moments? You say momentum, I dont know what is momentum in this contex but if it is the moment descriptives you talking about then you should use several of them to identify a pattern. Moments are descriptive quantities that can be used to represent patterns to some extend. But you need at least 7 or more moments for each pattern. Also there are different type of moments, Hu's moments could give better results than standardized moments.
I assume you use a scripting language to compute. You would go nuts with C++ or other low level tools.
#7
Posted 07 January 2007 - 02:54 PM
#8
Posted 08 January 2007 - 02:55 AM
#9
Posted 08 January 2007 - 10:47 PM