tag:blogger.com,1999:blog-5560052943772419320.post6675941696509710755..comments2024-02-28T11:21:54.924+01:00Comments on Dekalog Blog: Modelling Prices Using Brownian Motion.Unknownnoreply@blogger.comBlogger6125tag:blogger.com,1999:blog-5560052943772419320.post-63971443194714102012013-11-16T18:30:37.327+01:002013-11-16T18:30:37.327+01:00Piotr,
Rather than publish a mathematical express...Piotr,<br /><br />Rather than publish a mathematical expression of the method in a forthcoming post I shall give C++ .oct code for this idea. The standard C++ looping structure might be easier to follow than vectorised Octave code.Dekaloghttps://www.blogger.com/profile/07016889838486083128noreply@blogger.comtag:blogger.com,1999:blog-5560052943772419320.post-18752533381877192532013-11-13T10:57:07.181+01:002013-11-13T10:57:07.181+01:00I tried to implement with different programming la...I tried to implement with different programming language but I failed. I can't understand well Octave code. The description above looks a little different then code, but I am not sure. Would you please be so kind and publish a mathematical expression of your method?Anonymoushttps://www.blogger.com/profile/11731073922054484721noreply@blogger.comtag:blogger.com,1999:blog-5560052943772419320.post-23655144555508905982013-11-13T00:46:52.095+01:002013-11-13T00:46:52.095+01:00First congratulation for running this blog. I read...First congratulation for running this blog. I read it from the beginning till end and really gaveme a lot of inspiration<br /><br />About me. Seems we follow similar paths. First I started with DSP methods base on Ehlers publications, than after extensive evaluations I switched to AI methods, I'm an author of this thread<br /><br />http://www.trade2win.com/boards/trading-software/105880-3rd-generation-nn-deep-learning-deep-belief-nets-restricted-boltzmann-machines.html<br /><br />So my few comments which maybe can be useful.<br /><br />NN classifier. Why not to try more advanced classifiers specially designed to analyze big data file like e.g. pegassos or libocas. In my case I just developed interface MATLAB-WEKA to have access to all<br />WEKA algorithms, in my thread I compare performance like 10 algos at the same time.<br /><br />Evaluations methods. I believe to evaluate classifier you need to generate full confusion matrix to not only accuracy or precision measures as in this case you really don't know what is a rate of false positives. Also<br />ROC and AUC curves can be useful.<br /><br />Crossvalidation. Are you sure that this method is correct for time series evaluation ????? In such case you introduce future leaks i.e. train on data which occurred after test data !!!! Why not to do Walk Forward tests to see real performance.<br /><br />Some time ago i posted WalkForward result of all ehler strategies here<br />http://www.forexfactory.com/showthread.php?p=4154734#post4154734<br /><br />In general this thread in forexfactory is worth to read, some guys with very high expert level in DSP were posting there.<br /><br />Beside this in my opinion just checking the errors of the averages/filters is a bit disconnected from reality as it don't show how those indicators will work with the strategy<br /><br />As for decomposition of the prices. I 100% agree they must be decomposed somehow. Some method what you didn't mention<br />are like:<br />Stationary Subspace Analysis (SSA) (decomposition to stationary/non stationary parts)<br />MATLAB toolbox http://www.user.tu-berlin.de/paulbuenau/ssa/toolbox/<br /><br />Singular Spectrum Analysis (SSA) info everywhere<br /><br />Empirical Mode Decomposition EMD (Hilbert/Huang)<br />http://rcada.ncu.edu.tw/research1.htm<br /><br />fractiona Brownian motion is possible to generate with different values of Hurst exponent using Wavelet MATLAB toolbox.<br /><br />Hope it helps.<br /><br />Krzysztof<br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-5560052943772419320.post-13720800948747451302013-11-06T16:19:09.322+01:002013-11-06T16:19:09.322+01:00I also really like this idea for use as a trend in...I also really like this idea for use as a trend indicator. Perhaps built into an existing system as a trend filter.Anonymoushttps://www.blogger.com/profile/11869334055881673416noreply@blogger.comtag:blogger.com,1999:blog-5560052943772419320.post-10827392332073725902013-10-28T21:38:42.325+01:002013-10-28T21:38:42.325+01:00Thanks for the kind words.Thanks for the kind words.Dekaloghttps://www.blogger.com/profile/07016889838486083128noreply@blogger.comtag:blogger.com,1999:blog-5560052943772419320.post-42706976732866460202013-10-28T19:23:35.668+01:002013-10-28T19:23:35.668+01:00I really like this idea as an indicator for strong...I really like this idea as an indicator for strong price moves. Time to write some code Thanks! <br /><br />P.S. your blog is awesome keep up the good workAnonymousnoreply@blogger.com