Fractal Monger Reloaded

You probably wouldn’t know it but I have actually been extremely busy over the past few months resurrecting a previous creation of mine. It all started about two years ago, in early summer of 2014, when I wound up diving head first into more advanced topics such as machine learning, artificial neural networks, support vector machines (SVMs), as well as fractal geometry of financial time series. If you’re not familiar with most or any of these subjects then I wouldn’t be surprised as it’s pretty dry material, especially if you aren’t naturally blessed with a mind for advanced mathematics. For the record – I clearly am not. Which is why my absorption rate was rather slow and painful. But in the end brute determination and a stubborn disposition once again trumped my humble intellectual abilities and I was able to wrap my mind around the basic concepts.

Unfortunately however I was unable to reproduce the results of any of the most promising white papers I had collected. Perhaps I was missing critical information but most likely I had simply screwed up or misinterpreted some essential detail. Which is what happens when mere mortals aspire to soar toward the sun with nothing but a contraption made of wax and feathers. I have however been known to compensate for whatever I may be lacking in other areas with a healthy dose of imagination and resourcefulness. Which is what eventually led me to Fractal Monger.

Instead of using neural networks or multi-dimensional projections via SVMs I resorted to a statistical brute force approach aimed at determining fractal behavior in financial data series. The two approaches may be related by while the former leverages dimensional projections the latter simply focuses on data volume in combination with improving symbolic approximation. My efforts resulted in the first version of Fractal Monger which looked promising but in the end failed to deliver convincing results. Having already spent over six months on that project and wishing to go on with life I then decided to put the whole affair on hold and perhaps revisit the topic later.

About three months ago some of my ongoing work related to Scalpius (i.e. a short term scalper) led me to take another look at what had now become an ancient project, most of which I had almost forgotten about. My theory was that I may be able to use some of those fractal patterns in order to correlate short term trending behavior in various stages of volatility – the latter of which I am already able to predict quite effectively. Maybe due to a refreshed mind when reacquainting myself with my old code I quickly realized some fundamental flaws in my prior approach and set about correcting them.

In the process of testing my new approach I had a bit of an epiphany which in essence boils down to my realization that any inherent fractal behavior in financial time series may actually turn out to be multi-dimensional. At least empirical research in large number of scientific time series and geometric natural formations seems to point toward that assumption. In other words, instead of looking for the very same fractal on only one level, I decided to start parsing for them on multiple timeframes/levels. The exact details would bore you to death and I won’t hazard to test your patience. But suffice to say that the new approach yielded higher ranking fractal behavior across all symbols I tested. Due to a clever labeling scheme and the use of Google graphs I even started to recognize similar patterns/fractals across completely unrelated instruments, which is extremely promising for obvious reasons.

The road forward is two-fold. For one I decided to resurrect the Fractal Monger alerts which I had briefly offered back in 2014. As of right now I am able to send email, jabber, and twitter alerts but am leaning toward the latter two. Reason being is that the final settings and symbol selection may result in a flood of messages that would exceed some people’s noise threshold or land my domain on overly sensitive email spam lists. Twitter is a great network for delivering these types of alerts, with the only downside being that if I ever decide to go the pay2play route then I will have to figure out how to send directed messages via my REST framework. That’s a challenge for another day and one I would only consider tackling if I see Fractal Monger gain more recognition and hence subscribers. Jabber alerts are best IMO but you would have to get set up with an XMPP client on either your mobile phone or computer. If anyone’s interested I have put up a page with instructions on how to get set up.

The second avenue for this is of course fractal systems development. I may be able to make some enhancements to CrazyIvan or Scalpius but won’t really settle on a particular course until I have learned more about how fractal patterns distribute on various time frames and whether specific market phases are more conducive to the occurrence of fractal behavior in financial data series. What you can be sure of however is that this is only the first step in what appears to be a very promising and hopefully profitable direction.

Before I forget it – you can now sign up for Fractal Monger if you wish to receive email or Jabber alerts. Alternatively you can also hook into my twitter feed, which is probably the easiest route (and also free of course). If you go the email/jabber route then be advised that I will most likely disable the email module once I add two or three symbols and then only deliver alerts via jabber/xmpp and Twitter. But if you are interested then I suggest that you sign up for now in order to evaluate the signals. You can always unsubscribe or shoot me a message to be removed from the list.

One last thing – this is early beta – so please do NOT trade on any of the signals. They should simply be considered to be edutainment as of right now  By the way – any input will be welcome – I serve at the pleasure of this community and as I’m operating in my own bubble I would very much appreciate third party insights, especially if you had prior pertinent exposure to the subject matter.