Introducing Fractal Monger
In the past few months I have been immersing myself into the world of machine learning with a strong focus on data cross-validation. Quite frankly intellectually it’s been a brutal toil for me as my math skills had gotten a bit rusty since college and since this was completely new turf for me. I will spare you an account of all the trials and tribulations I have been through – as you can imagine this subject is not exactly considered light reading.
Repeatedly I found myself researching a particular promising avenue for weeks on end which in the end turned out to be nothing but the precipice leading to a steeper mountain of work hiding behind it. However without understanding the basic concepts the more advanced ones will simply confuse or elude you and the premise of their evolutions will not be clear. As I spent weeks absorbing a mountain of white papers I often only realized their significance (or lack thereof) many weeks later. As I started to put theory into practice some of the ideas I had developed over the years were confirmed – the ones which weren’t living up to the data were discarded.
At this stage I would not consider myself a quant boy of any sort but as they say I do understand enough to be dangerous. Given that, what I’m showing you today is only a first foray into developing more advanced systems and pertinent tools. And it addressed an issue most short term traders engage in at some point in their career: time series fractals. It’s the one aspect of technical analysis that has always fascinated me but which I never really could put to work.
When it comes to historical precedence most of us spend our time driving blind – we see a time series and we don’t have any idea as to whether it has occurred in the past and/or if it has any statistical meaning. Which doesn’t really matter if we have a functioning system that has shown an edge – perhaps that system actually implicitly takes advantage of one or more time series fractals. To be clear – I’m not talking about candle patterns here although one could easily investigate fractals on that basis. As a matter of fact it’s something I may do in the near future – but it’s not what I have been working on.
This is what I put together over the weekend – it’s based on a converge of several fractal categorization approaches I have been working on. In a nutshell what it always boils down to is this: What are the odds of the next candle closing higher or lower? Well, in this case we actually know:
The current fractal has occurred 385 times in the past.
It ranks as the 45th most frequent fractal during the past 50000 bars.
The next candle has closed higher 171 times, lower 206 times, and equal 8 times.
Frequency of occurrence: 0.09%
Higher/lower ratio: 0.8
Lower/higher ratio: 1.2
That is the email message I sent myself for testing. It wasn’t as pretty as I’m still futzing with HTML formatted emails but you get the idea. What’s extremely cool here IMO is the fractal graph I’m generating on the fly. As you can see it’s based on four closes, which seems to be an optimal vector size in terms of frequency/occurrence and probabilities. The accompanying statistics tell me that the odds were pretty mixed on this one and that it had a low frequency. It did in the end close lower but I would not have traded this information.
The current fractal has occurred 617 times in the past.
It ranks as the 13th most frequent fractal during the past 50002 bars.
The next candle has closed higher 272 times, lower 316 times, and equal 29 times.
Frequency of occurrence: 0.03%
Higher/lower ratio: 0.9
Lower/higher ratio: 1.2
This is what I got an hour later – we wound up closing higher – again the statistical odds were below my threshold which is currently 1.5 (either on the H/L or L/H ratio). And that means outside of testing I would not send myself (or any of you) this fractal as trading on such low odds is tantamount to a coin toss.
But for the next week or so I would like to test it on an hourly basis without the threshold filter, meaning there will be one email per hour on the ES and the EUR/USD – you have been warned. If you like the concept/idea then feel free to sign up for Fractal Monger in the usual place, it’s free right now.
Over time of course we expect to run into fractals with higher statistical value – I actually don’t think we’ll get more than one or two in a week, even if we run it on two dozen symbols. But having done my homework I already know that some of them have very high odds and could easily tie into systems or discretionary trading. Let’s see if we can catch some of them.
Your input/comments/impressions are of course highly encouraged. See you later today.
It’s not too late – learn how to consistently bank coin without news, drama, and all the misinformation. If you are interested in becoming a subscriber then don’t waste time and sign up here. The Zero indicator service also offers access to all Gold posts, so you actually get double the bang for your buck.