All Things Statistical - page 7
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Multicollinearity Part 4
Good Day All,
Apparently, there are quite afew more ways to correct for multicollinearity which i have missed out:
Combining variables
Differencing
Data Transformations (which has alot of pitfalls especially...)
https://analysights.wordpress.com/2010/07/22/forecast-friday-topic-multicollinearity-correcting-and-accepting-it/
Apparently, it seems fairly common by alot people to use PCA or penalization to resolve the multicollinearity issues. A study below seems to come to the conclusion that simple methods work as well as more complicated methods, noting that changes in collinearity structure is possibly devastating for all adjustment methods.
Collinearity: a review of methods to deal with it and a simulation study evaluating their performance - Dormann - 2012 - Ecography - Wiley Online Library
Anyway, coming back to the collinearity issues with the usual TA indicators.... Very very high collinearity issues involved here as seen below. One thing i noticed here is that Momentum with CMO is 0.93 while that of CMO with ADX is at 0.82, which would fairly likely mean Momentum and ADX probably will have a high value too, when ADX is classified usually as a Trend indicator... Oh well.. Too much collinearity even amongst indicators from different categories....
Wintersky
Cheers
Multicollinearity Part 5
Good Morning All,
Been looking at practical instances of avoiding Collinearity and having the outlook for many years that alot of stuff around are largely collinear. One example of something that caught my eye lately:
Trend-spotting @ Forex Factory
The rules are:
If the bar closed UP, then you look at the HIGH of that bar and you count how many bars it has been since any other bar has been that high.
If the bar closed DOWN, then you look at the LOW of that bar and you count how many bars it has been since any other bar has been that low
The author of that idea at FF has a really interesting idea and someone had also coded out the indicator there for such bar counting method (i know right... this feels alot like Demark's TD Sequential type of stuff).
What's interesting to me here is that this really offers a different way of counting information instead of the old usual fixed 20 bars or 30 bars indicator parameter, as the bar count under this method can reach to infinity possibly. So this is practically certain to not have any collinearity issues.
Low Collinearity= More Edge possibly
The only other thing here is if it offers a edge. Interesting certainly, but i feel that some of the problems here are quite glaring. This would certainly be considered a breakout system of sorts.
Wintersky
Cheers
Mathematical Proof For Trend Existence In Financial TIme Series
Mathematical Proof for the existence of trends in financial time series for those who do not belong to the Dark Side yet...
May the Force Be With You!!!
trendproof.pdf
Wintersky
Cheers
Approximate Entropy For Various Currency Pairs Daily Data
Good Day All,
Found this thing here and looks pretty interesting. Approximate Entropy. Supposedly its similar to entropy but with better applications to noise situations.http://en.wikipedia.org/wiki/Approximate_entropy
Below is the currency pairs on daily data with approximate entropy analysis. Suggested Pairs to trade are USDJPY and EURUSD... Would be nice if someone could find the same for intraday data and make an Approximate Entropy indicator for MT4 (im a non-coder sadly)
Kindly share with me if someone manages to code it! 
Determinism and Entropy: Choosing which Forex pairs to trade using R | Mechanical Forex
& below here an Entropy adjusted Random Walk. Higher amplitudes (Probabilities) are observed and data is largely contained on a lower spread of returns...
http://www.phys.sinica.edu.tw/~socioecono/econophysics2010/pdfs/ZapartPaper.pdf
Sadly being a non-coder, i'll probably have to find a way to find something to substitute a similar function or something.... Approximate Entropy seemingly looks more interesting than basic entropy...
Wintersky
Cheers
random walk on the wild side...random wok (chinese)...will figure out something else as time passes bi...
Approximate Entropy Part 2
Afew more papers below on Approximate Entropy:
yentes_abme_2012.pdf
http://www.pnas.org/content/88/6/2297.full.pdf
Personally as i look further into the characteristics involved here, afew things seem to stick out. it's not meant to be able to distinguish noisy/perturbed data and in the case of heartbeat data it uses r= 0.1 or 0.2 of the SD of a specified data period length. But in the financial side with heteroskedastic variance +variance being especially wide....
& then theres the part about 1,000 or 2,000 data points to achieve convergence as shown below.... would pose too much of an issue for usage on a real-time basis...
i dont know what to think about this. Feel free to comment. Would be interested to hear more about this.
Wintersky
Cheers
Thank you Wintersky for bringing these studies to the forum. Right now I am so busy doing the regular technical analysis that seems to be working for me that I currently don't have time to develop a system with these statistical methods. Maybe if I live long enough to win the lottery or retire from my day job I can spend the time to develop a statistically aided system.
Again thank you for your posts. Keep them coming.
Good Day All,
Found this thing here and looks pretty interesting. Approximate Entropy. Supposedly its similar to entropy but with better applications to noise situations.http://en.wikipedia.org/wiki/Approximate_entropy
Below is the currency pairs on daily data with approximate entropy analysis. Suggested Pairs to trade are USDJPY and EURUSD... Would be nice if someone could find the same for intraday data and make an Approximate Entropy indicator for MT4 (im a non-coder sadly)
Kindly share with me if someone manages to code it!
Determinism and Entropy: Choosing which Forex pairs to trade using R | Mechanical Forex
& below here an Entropy adjusted Random Walk. Higher amplitudes (Probabilities) are observed and data is largely contained on a lower spread of returns...
http://www.phys.sinica.edu.tw/~socioecono/econophysics2010/pdfs/ZapartPaper.pdf
Sadly being a non-coder, i'll probably have to find a way to find something to substitute a similar function or something.... Approximate Entropy seemingly looks more interesting than basic entropy...
Wintersky
CheersI'm not good at math, better with language. Can you convert the entropy calculation into comprehensive language (english) ?
Thanks Wintersky for the approx Entropy links. I have the correct packages installed in R but I did not get the code to run just yet. I might just need another coffee. I have also found similar code for MatLab which I will look at later as well. If I get either to work properly, I will post some results.
Here is the link for further details although there does seem to be a couple of different flavors of this if I look at the various packages for MatLab. I think for intraday use with 24 hour markets, it would be necessary to use non time based charts to better capture price movement.
Approximate entropy - Wikipedia, the free encyclopedia for a summary.
Tanzenkran: Thanks. Sure will keep the posts coming, even though i believe most of the stuff has been known and researched to death by some of the outstanding researchers and traders around in this forum. i dont have a coding or mathematical skills sets foundation, so this is my way to contribute to this forum as much as i can as i've learnt alot here and still need to learn even more from the many great people here
Airquest: I'm not good at the maths too. So what i do is read the theoretical interpretation in all the files i see 1st instead of the maths. Read the files i've linked/posted above. The wiki link as stated by hugesfleming seems good too, though some of the maths i dont recognize too. I'm also now starting to really read the PDFs too as i only had a brief look when i posted those as i was in a rush for time to go out then.
You might want to look at SampleEntropy too as it has some crucial benefits as compared to ApproximateEntropy. One of the files above was talking about the former being better than the latter too, especially about the "M" variable having data length independence...
Sample entropy - Wikipedia, the free encyclopedia
"ApEn measures the ‘likelihood that runs of patterns that are close remain close on next incremental comparisons".
Hugesfleming:
Thanks! Would be interesting, though im already thinking that as the article says, there would be alot of sensitivity to the parameters. The specification need for M and R variables is going to be a headache in real life.
Too sleepy to talk more. More when i wake up. Goodnight All!