Monthly Archives: November 2013

Substring Sequencing


Okay, here is the first sub-sequencing I’ve done. Starting with the basic one, the 4 string.

On the left is all of the 4 string sequences, contained within each day. In other words, sequencing re-starts each day; the max string is 24. The most basic sequencing is 1-2-3-4, then 2-3-4-5, 3-4-5-6, etc. I call this pure linear sequencing. It helps answer the question: if the current sequence is Up, Up, Up, what are the odds that the next bar will be Up versus down? This can be made more complex like I’ve said in a past post that by including different ways to count such as 1-3-5-7, 3-5-7-9, 11-13-15-17, etc. This can be expanded as long as the subset can hold it (in my case, 24).

My data, instead of Up hours versus Down hours, is Movement versus non-movement. Movement is defined as a new HH or LL, with the same idea that I’ve been using for quite a while now. HH/LL has to meet the context that it breaks the current daily High or Low.


On the left, the 4 string sequences for ALL subsequences. Linear, 1 space, 2 space, 3 space, etc. The data shows that in all sequence types, probabilities favor non-movement, aka an “X”. However on the right, I’m showing standard pure linear sequencing. Here it appears that there are a few cases where “M”, or movement, aka, a new Higher High or a Lower Low, is favored. Interesting.



Wave Modeling 1.5

As a small aside note, I do intend on building this wave thing up to something like wave modeling 6.0 someday. So each improvement warrants it’s own post.

Here we have the waves broken down after I’ve already broken them down to most of the movement. Currently A/B/C/D waves under my new filters account for 98.12% percent of days.

Each wave has two types: the UP version and the DOWN version. I didn’t have a great way to show each type without too much clutter, so for example, Type 1 of the A Wave is H. This kind of day moves only up, and never breaks the low. ┬áType 2 of the C wave for example is HLH. This kind of day makes a Higher Higher, then a Lower Low, then another Higher High that breaks the original high of the day (after including the new filter definitions).

The second chart shows these numbers as percentages within the waves. If I know for certain that an “B” wave is occurring, the chance that a Type 2 wave is occurring is 53%. They are all more or less 50-50, with the exception of the D wave which doesn’t have much data to begin with.

The third chart shows these numbers as percentages of all the data. The Type 2 of the B wave occurs 22.24% of the time.


Edit 02/14/2014:

Cleaned it up yet again. I really don’t mind redoing statistics since I almost always end up getting cleaner and more accurate statistics. This is my best yet:


Any other statistics have not been redone.

I figured the following idea would fail, but it never hurts to try. I tried to look at the minimum percentage that a wave will retrace to on average. What if the first retracement on an “A” wave commonly retraces to something between 20-30%, whereas the first retracement on a “B” wave commonly retraces to something like 50-60%? Thus, but just seeing where current min retracement is, it could give you a better idea of what kind of wave is occurring.

A careful look will show that the “A” and “C” waves show Type 1 first, while the “B” and “D” waves show type 2 first. This is because I column sorted it for net price moves. A type 1 “A” move is net up move, while a type 2 “B” wave is a net up move


The first number is the Average percentage, the second is the Standard deviation. Given the high numbers on the STDEV, I don’t think there’s much here yet.

Here is the same picture with the “fixed” type 1 versus type 2 to better answer the question posed earlier.

attempt2It appears that on the first retrace, the less complex waves retrace to a higher percentage, or in other words, the more complex waves retrace deeper. If the STDEVs weren’t so high, there’s a lot to look at. Even with such variation, there may be something useful here.


Wave Modeling 1.4

Added a filter on. The percentages are becoming more clear:



Total of 1503 changed bars. In other words, about 45% of bars go straight from HH to LL or from LL to HH (with no retracement)at least once. 583 of these bars make this change more than once a day.

Most importantly is to always maintain a tradeable edge. In other words, the filters that are to be put on can’t be overly complicated such that a trader or EA can’t actually execute them, or require future information. It should also present a decent number of trading opportunities and most importantly, create windows where trades can actually be executed. There’s no point, for example, of knowing that an “A” wave will occur if it takes 20 hours to determine that result.

Wave Modeling 1.3

Keep digging. I have a couple of steps to work through and I need to think about how the pieces could possibly fit together, but the idea is that after these next couple of parts, I can bring in the parts I talked about in my journal.


This data I took from a different dataset, so if one studies Wave modeling 1.1 and 1.3, they will notice the numbers are slightly different (like total days). Here, the “A” wave, where price goes in 1 direction and 1 direction only, occurs 11.26% of the time. Within this set, Price only makes 1 HH and then is in trapped between the open price and the high of the day 8 times . The average of the first retrace is about 44%. NOTE: this is the average retrace of the first pullback on every “A” wave, not just those that have 1 HH and only 1 HH.

digThere’s a lot to do with this data here, and additional manipulation that I think I’ll need.