Monthly Archives: May 2014

Daily bars within the week

Don’t think I posted this, but I think I did it before in a very ‘what ever lets see’ type of fashion. Importance of archiving! Stashing a bunch of random, general, information about the market is generally not useful, but when you have a specific need for it, it’s nice to have.


Up weeks on the left, Down weeks on the right. More or less the same thing. Things to note: 4 days of the week are more common and 2, and all 5 days moving in the same direction is more common than 1. Expected, but nice to have numbers on it.

The engine verified (more or less) the stats I observed in the last post. My first edge! Landmark day 😀

What I’ve managed to pull off, for the time being, is a roughly 80% non-losing rate. This is very different than a 80% win rate. I think I’m losing about 20%, winning about 50, and breaking even/slight loss on the other 30. However, the gains are still too slow. Given it’s extreme fill requirement, I’m only generating about a trade every week. Engine showed about an 8% gain from Feb 1 to May 20th of this month. Nice to start, but a lot more work to do. Initially I thought to drop the time frame down to generate more signals and more trade opportunities, but then I thought, what if I bring the time frame up? Having a weekly signal would actually allow me to trade every day, after some more back work is put in.

Direction Trajectories Pt2

Perhaps a bit pessimistic, although I think it’s being more thorough and careful that anytime I see good results, I think there’s some error in calculation. I’ve looked over the data and can’t seem to find anything wrong with it. (step 1). Next I’ll code it into a macro engine, and I’ll really know if something’s wrong with it. If there IS something wrong, I do have an idea about where it came from. If not, the beta engine should be rolling out pretty soon. Hoping for the best; I do feel like I’m getting close to hitting something that really works, but it could be a couple months out still.


Atm, consistency across the board on these stats is really what’s bothering me; it doesn’t quite make sense. The prelim engine shows similar results, even with a dummy signal. I think though, that if the rest of the code is correct, it’ll at least show me something about the.. fractal(?) nature of the markets and a hint about how to use 50-50 edges to ones advantage. OR. I’m just seeing things and I’m way off base. D: We’ll see where the engine takes me.

Update on direction

I’m failing hard lol. I don’t think it can be quite done with the knowledge that I have now, and the scope I’m using. I’m trying a slightly different approach, which, if I can find an edge in it, will lead to a “complete” strategy quite easily.


This shows some stats for the rebuilt projection lines; One for up one for down. Using reverse cumulative frequencies, it looks like once price has hit the level, it’s good for another 50% of the level a safe (75%) amount of the time.

A look at how to build deeper metadata

“We must go deeper”-that one movie.


I haven’t completed a statistic for it, but figured I’d post about it so that I can get a better grasp myself. I believe 7th once said that the system in it’s end phase was so complex that he couldn’t even describe it. Given that the formulas are stored->spat into a new formula->stored elsewhere->used as another input in a formula, etc. and so on until a final output is created, it makes more sense to me now; it’s surprisingly easy to forget what it is I’m actually measuring.



I read somewhere that 7th referred to the elements of DV as the following:

H/O: Leading Long

O/L: Leading Short

C/L: Trailing Long

H/C: Trailing Short

Now, you can take the 1 dimensional delta and expand it by attaching it and fusing it with the past, like turning H-L into H-pL and pH-L.

Doing this with the 4 deltas (not the Omega delta) creates a total of 8 new deltas.

H-O becomes H-pO, Ph-O, etc… If H-O is the leading long, this would turn the present into it’s own sort of long trajectory, and the past into the short trajectory, aka:

current high-previous open=leading long (long)

previous high-current open=leading long (short)

So at the top I’ve relabeled them as such: LL, LS, TS, TL, each with an L or an S to create the 8 deltas.

When it comes to messing with the data, there’s a lot to do… A glimpse can be seen on the far right of the Screen shot. For example, on the far bottom right, the section labeled SL, LS, and TSL, TSS.

This table translates to looking at the Trailing Short Long and the Trailing Short Short, in regards to 1 bar compared to the next. SL stands for Short to Long, and LS stands for Long to Short. Thus, the mini stat at the bottom is saying when a day moves from Down (short) to Up (long), the Trailing Short Short is larger than then Trailing Short Long about 68.2% of the time. 1 Bar statistics isn’t very useful for forecasting, but if you forecast multiple bars in advance…

Trying out some version of the “see how much is too much” idea. This doesn’t even take fills into account anymore, and I haven’t figured out a way to incorporate them in yet.

Second glance fills

I wouldn’t be surprised if a fully functioning system was built off of fills and only fills. Something that I’ve noticed about them is how great they are at following price.


This is a stat I ran to basically determine how often the under or overfill is the larger/smaller of the HO vs OL. For example, if the H/O delta is in deep over fill, it will be the larger of the two deltas 98% of the time. so on from left to right; HO over, HO under, OL Over, OL under (%’s adjusted to reflect calculation).

Some obvious pros and cons. The better it follows price, the more it behaves the same way price does. Fills do have extreme limits though, and that’s what I think the brilliance in it is.

Now, the current issue is that just because something is underfilled to the extreme, while there ARE probabilities that favor it filling up (when filtered correctly), I do NOT have the ability to compare it to the other end. In other words, If I see HO underfill down to 10%, I can bet that it will fill in more in the next period, but I can’t be certain that it will fill a point where HO is now greater than OL.


Taking the rule of thumb, looking for 60%, preferably 65-70 as a base point.


Looks good until I ran the back stats to show that just because it’s filling >80 or 100%, doesn’t make it bigger than the other side. Aww.

Direction Trajectories


Small chunk (~110 days) but it actually looks pretty good I think. Real price on top, Trajectory on the bottom.


Edit: Instead of making a new post I thought I’d just add on to this one since it’s so short.

The above picture looks great, the problem as suspected really was the sample size. Basically the trajectories are a good representation of where price is going, but only sometimes. Typical. Which line over 0, over the long run, does not help predict next bar C-O. Seems hard to believe given the first set of pictures.


This next set is still decently good, but no where as good as the first one. There still might be something here, but I’ll have to add in some more elements to filter out some things.


Initial movement compared to daily MM

As I try to pin point and search for ways to track the 4 keys to trading according to 7th, they all start to become mixed together. It’s suppose to be that way, and not at the same time. It’s hard to explain. I worked on some stuff yesterday that I realized is very similar to saying something to a conclusion I came up with earlier, working in a different area.

I had a small idea yesterday that I figured I could execute fairly easy since the base part was in code I already had. IMO, a day can have more than 1 major move, or trade-able move, but if we were to confine the day into have just a single direction, the way I’ve done it using the old wave analysis is correct. So, I wanted to look at a simple question that I’ve kind of subconsciously been working on. When I was working on A waves in the original framework, the question I had to solve (because A waves made up such a large portion of the dataset) was: At what point is the A wave more likely to be an A wave, as opposed to any other wave? I’ve kind of solved this by answering: If price has moved x pips in y direction, how likely is it that the MM is also in that direction?


For example, if price (relative to open) is up 20 pips, the MM is likely to be Up as well about 57% of the time.

Surprising results..