All Things Statistical - page 4
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
That your hero good old Pava?
Good point
Once when I threw away the perfection expectation I have found out some real gems here
Technical Analysis Validity
An Interesting/Daring Paper by Lo, Mamaysky and Wang on proving that Technical Analysis is not all voodoo science but has significant effectiveness in investing. About time that a good look is taken at Technical Analysis. Enjoy!
http://www.cis.upenn.edu/~mkearns/teaching/cis700/lo.pdf
Wintersky
Magic Wang...2 Wangs don't make it right...Lo and behold...(Mamaysky курган)
Magic Wang...2 Wangs don't make it right...Lo and behold...(Mamaysky курган)
No running Pava since you're here! Give some value to our statistical discussion here or something. Utter the magic words!
o'k...here you go: "There is no coincidences"
No running Pava since you're here! Give some value to our statistical discussion here or something. Utter the magic words!
Effective Sample Size
Hi all,
Been reading up on ESS issues. According to William Echkardt's viewpoints, kurtosis issues demands a large sample size in the wide hundreds. From a different angle, the presence of autocorrelation also demands an increase in sample size...
Here, the graph below describes false Type 1 Rejection rates under a Standard T-Test, 100 samples each from a Gaussian distribution.
Statistics and Climate – Part Four – Autocorrelation | The Science of Doom
Another Paper below at page 8 shows describes ESS requirements = 3 x Sample Size Used when Autocorrelation is at 0.5.
notes_3.pdf
Some studies state that 1minute information in the FX markets suffer from very significant Negative Autocorrelation effects, probably due to Market Microstructure Effects. & the corresponding kurtosis increases too at such timeframes... What sample size really would be enough? & are there any simple generic across-the-board correction values for AC for almost all/all forms of indicators?
Feel free to chip in your comments
Wintersky
Autocorrelations
Significant 1st Order Autocorrelations show here for Daily Returns.
Here, significant first-order Negative Autocorrelations at 5 minute time series at 1% significance for multiple pairs. 2nd to 4th order effects were also significant too, though with much less magnitude.
Would it be incorrect to equate Autocorrelations in a way with Momentum/Inertia?
Wintersky
Additive White Gaussian Noise. SNR= 27DB. Here, signal extraction by EMD is degraded badly due to noise characteristics. My personal feel is that it's nonlinear characteristics brought about by the splines fitting is an issue here.
http://danielkaslovsky.com/KaslovskyMeyer-EMD-postprint.pdf
& here below, the Morlet Wavelet (from Huang's original Hilbert paper) shows end effects issues at about 10% each on both ends of the data. Probably, a similar extent can be said for both EMD and SSA as a general usage guideline.
http://tec.earth.sinica.edu.tw/research/report/paper/20070711HHT.pdf
Wintersky
FX Tick Volume Correlation With Trade Volume
A study done using Pearson's Correlation with different signal providers at the hourly basis shows 90+% variance in accounted for, with R-Square at 90+% too.
Additionally, as to whether it is applicable for lower or higher timeframes than the above is up to further debate and testing.
Wintersky
fxvolume_tick_volume_vs_real_volume_study_1.pdf
Time Frame Considerations
Statistical mean of the market [quant corner] - Page 8 @ Forex Factory
Proximus's study (from Forexfactory) showing a swamping of noise characteristics at TFs below 60mins/240mins, while Old Dog's study below based upon Mutual Information shows a similar conclusion.
The Optimum Time Frame for Trading @ Forex Factory
Above here, Dacorogna's study finds that significant negative autocorrelations occur up to the first 4 minutes in foreign exchange, with 1 minute timeframe being extremely negative.
==============================================================
Reflections:
In negative autocorrelations time series, the mean would be a good way to gauge the sample population characteristics. But the above graphs shows some kind of "contradiction", with median usage being more efficient than means usage in lower timeframes, even during pupported zero/insignificant autocorrelation timeframes. A possible conclusion here might be that the problematic characteristics come from more than just autocorrelations, possibly due to distributional kurtosis?
I dont know...
Wintersky
Cheers
