Why is your correlation matrix "dirty"?
What is going into it, and how do you intend to use it?
What is going into it, and how do you intend to use it?
Optimized Risk vs Reward Equation 335 replies
Loss Less Optimized Strategy 63 replies
Optimized Parameters in Metatrader Testing? 3 replies
"Optimized Parameter" feature in EA? 0 replies
DislikedJaguar, please explain how do you intent to use this software.
You may check also this just for information.
http://www.jurikres.com/faq/faq_wav.htm#top
This also looks interesting
http://www.jurikres.com/faq/faq_ddr.htm#top
However I am interested how can I do similar things for free with the free available software.
S!Ignored
DislikedIt certainly looks like DDR is some derivative of PCA. (Which is a lot like SSA). You can get free MatLab software to do PCA. Just Google it. Also most neural net packages a la NeuroShell have PCA and wavelet add-ons. Alternatively you could find C source and make a DLL.
Usually both PCA and Wavelet transforms are block transforms requiring the transform to be recomputed each new sample, however there are forms of PCA that can be computed incrementally. Google CCIPCA if you are interested. The IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE,...Ignored
DislikedMadCow,
Can you attach the DLL file for CCIPCA ?And also can you give the link for incremental SVR algorithm?I could not find it in Montana's web page?Cheers.Ignored
DislikedCCIPCA
*********
Well, everything reachs its own limit soon or sooner
Here, in the case of the CCIPCA, the higher-order engeinvectors
converge, since their properties of convergence are based
on the fact, the previous eigenvectors have been perfectly calculated
So, there are 2 hidden traps
- What to do w/ small eigen values ? I wrote this in my previous post
- what do you expect if the previous eigen vectors were not
perfectly calculated ??
So, this theory build something wrong upon wrong estimation.
There is a paper, indicating a new direction,...Ignored
DislikedAs for me I use SVM implementation by RapidMiner. So far no success story. I tried a simple end-of day model with high/open/low/close with sliding windowing. It did not stop to counter the market during the high trend of the euro/usd. After more than 10 consecutive wrong predictions I was tired (Of course this was just testing).Ignored
DislikedAs for me I use SVM implementation by RapidMiner. So far no success story. I tried a simple end-of day model with high/open/low/close with sliding windowing. It did not stop to counter the market during the high trend of the euro/usd. After more than 10 consecutive wrong predictions I was tired (Of course this was just testing).Ignored
DislikedSo if there is a black swan you just cannot maintain the same amount of risk, you just cannot let a big stop loss and go away (end of day strategies are suicidal). The risk management should be different.Ignored
DislikedThat is why the real success stories for artificial intelligence models are by the guys who not only can operate a complex model but who can also be tuned with the market conditions. And unfortunately at our retail level I do not know a short cut for market knowledge and experience.Ignored
DislikedI use the VIX for this. Any blackswan produces a huge blip on the VIX. Just as long as it's underlying market is open.Ignored
DislikedFurthermore
As anyone here noticed, the market is changing, and moving from ranging to trending, but too, changing of speend during those major periods, I thought they were something else able to help us to identify in which period we were.
As John noticed, IA, sometimes deduct values, Generally, becoz the market is moving to anonther kind of velocity, prices, momentum, trend and so on, the IA results are not enough relevant.
Well, we were talking about matrix embedded dimension, correlation matrix for each indicator values, but in fact,...Ignored
DislikedWell, it seemed to me
- once, the market phases will be identified, a different strategy will be applied for each phase
Today, we are just at the beginning to split the market waves into several generic parts.
After, for each part, there is one corresponding Time Series, and a dedicated IA, after understanding the nature of the phase
After each time, the market is evolving and changing of type, a new and corresponding algorythm should be applied.
So, w/ this kind of philosophyn between # phases, (it should take a few bars), no trading...Ignored