# Rabbit Hole Series #2 extended

Some pictures from the H1

Can’t quite get the logic correct to get all my numbers to update live time, but I do have a good way to create these graphs historically to check multiple time frames, various h values, and the like. This expands the possibilities I can measure and cross check, and perhaps follow SBs advice on using an MTF approach. I’m not sure if excel is better suited for checking these things compared to MT4, but one certainly has to be clever in how to organize and pull the data to check for hard proof that these concepts actually work.

I currently don’t know which levels (70/30, 75/25, 80/20, etc) are better. I think to better capture these opportunities though, I want to expand the average length of each wave (more potential).

Giving clues on in-trend pull backs and end of trend exhaustion.
I’m noticing that the signal is usually slightly early; A long signal usually has a one more dip low, and a short signal usually has one more push up.
Enough cherry picking, time to try to find a scientific way to analyze all of this.

# Rabbit Hole Series #2

Trying to build a basic sort of momentum indicator.
I think I may have to revisit the premise of how I’m building it (when I can “see” it) but I’m in the baby stages atm.
The more tailored the tool is, the more wiggle room I have to make a tool that is one-dimensional. That is, normally if the tool needs to determine where price is going, then it needs to distinguish both long and short, as well as strength. But if I already know the direction, then I only need it to determine strength.
Specifically, when comparing the accuracy of an indicator that giving long signals, I don’t need to check a DR (down retrace)  against DT UR and UT, only against DT.

The blue line is a measure of C-PL, and the red line is a measure of PH-C. Then I modded the values using a sort of “strength” multiplier based on the TAC TCD (H-PL vs PH-L). A question that I was wondering is if price makes a move in which the bar prints a downward pinbar like so:

Should price be considered bullish or bearish? How does context matter in this case? If price is already moving down, do I consider this a “slowing bear”, and thus bullish? If price was moving up prior, do I now considering this new strong selling action, and this bearish? If the context matters, (which logically it should?..) that would mean that the strength multiplier needs to also be multiplied by.. the strength of the strength multiplier?..

Imagine it this way.. if bull movement is A, and bear movement is B, then the strength of A would be A/(A+B), and the strength of B would be B/(A+B) (that’s one possibility at least). But if the context of the current strength matters, then the strength of A would now be something like [A/(A+b)] * Y, where Y is equal to the slope or average of past strength.. or some other indicator would need to indicate the strength multiplier Y to create the current strength, to then be used to measure the actual strength of price. Oy.

..which one is which? I couldn’t really tell the difference. Accounting for natural bias I don’t think that if these were mixed together, that I could separate them back out. Normally I would expect (or hope to produce) that the blue line is measuring bull strength, and the red line is measuring bear strength. Thus whichever is on top shows the current dominant vector, which the other line fighting to overtake the current direction. It just doesn’t work though.  But there was something interesting that I missed the first time, which shows here:

What I noticed is that (after the 4th rendition), that the red line does a pretty good job of following price. Usually not something to be excited about, but I did manage to also get price shrunk into a 0-100 oscillator with a bell curve.

So disregarding the first couple points of each wave which are distorted due to having too few data points, I may be able to detect price extremes using it. The issue now is what to do with the blue line..It’s not perfectly inverse, so perhaps there’s something to look at in the spots where they differ.

Here’s one example of printing an extreme (over filling?) to give a short signal when price actually has plenty of room to move down.

# Trend is your friend?

Is this scientific proof to follow the trend and it’s strength? (pt 2)

I’m cautious to make more additions to this. I think at this point I want to be veryy critical and precise with improving at the core components. I’m kind of viewing this as “gen 1” of 1 part of a system of x number of components if we’re speaking about the SB metabrain style of trading.

The picture is pretty self explanatory given one is aware of what h values are. What’s interesting is the difference in strength between the low and high extreme h values.
For context:

Goal:

Currently (after nearly 18 months LOL):

The journey continues.

# Progress on Metadata and TCDs

SO difficult!

When I find myself studying metadata and the resulting TCDs, I find myself struggling with a lot of philosophical, chicken-and-egg, and catch-22 type of issues.

Creating additional metadata is easy. Quite easy. The issue is analyzing it correcting and discovering what is “right” and what can be improved upon. I suppose first I should define how I view research in general.

The goal of research (for me) falls into two categories: Signals, and system/knowledge.
Signals: each signal has an absolute end: does the signal work or not? Is the bar recurrent or transient? Discovering strong signals (ones that don’t fail often) is difficult, but calculating whether or not your signal works is easy. You can set it with either a time control or pip control (stop loss) and at the end, you get a clean number. x% wins, y% losses.
System/knowledge: This type aims to take the information displayed through raw data, and strip it down to only what is necessary: highs, lows, critical points, etc. Things that fall into this category are like all the wave models I’ve created or market states.

SB TCDs clearly fall into the second category, and the issue with these types of analysis is that it’s hard to know if I’m right or wrong.

Since TCDs are based on raw data, they update as time progresses. The net end of a TCD is some sort of output, such as a number or translated into “long”, “short”, “flat”, “expand”.
Now, If my TCD is signaling “short” anywhere in the green box, is the information “correct” or not?

Clearly, the chart moves from an uptrend to a downtrend, and clearly the leftmost side of the box is capturing the end of the bull move. If I’m working on a TCD overfill signal that is suggesting that the longs are overfilled and a short movement is necessary, at what point is the signal considered correct? Given the nature of the market, an overfill, whether it be long or short, is bound to be correct at some point or another.

My conclusion from this was that “correctness” needs to be in conjunction or agreement with another TCD. But this leads to similar types of questions. If my signal changes from Sell to Buy, when do I buy? For how long? For how many pips? What determines if the baseline TCD is an accurate representation of price? Initially I thought it was possible to use something like the standard H-pL and pH-L. However to use these effectively, it seems like it’s simply morphing the chart into another chart. (from the standard OHLC time based chart to a TCD connected time based chart.

To take the chart from SB’s thread with the quote:

“Based on just these questions alone – I could look at this one chart alone and tell you that this currency pair was about to make a move Long for at least 80 pips (the harmonic average) and that it was going to do it within the next 24 hours.”

My issue is that historically I’ve seen stuff like this occur with other TCDs, but it has issues that I have with this chart as well:

1. Period 44 looks like a similar position where the Long TCD looks like it will cross the MA, but doesn’t.
2. Even if the long TCD DOES cross over the MA, it doesn’t mean that it will cross above the Short TCD and it looks highly likely that the Short TCD will still be higher.

I can think of some answers to these issues, but only in more general abstract ways. aka, to deal with point #2, develop a signal or indicator that will indicate when the short portion will be over so that the long portion may run. I see this as the following:
Tomorrow the day (starting from open) will see 80 pips up and 80 pips down with 75% probability. Based on this, you cannot successfully trade it. What you would have to do is to develop a way to know (or reasonably know) when the short or long portion will be over to trade the other side with good RR. On and on! Systems need to be developed to monitor other systems to monitor other systems that give output on what will possibly happen next. I believe that this is ACTUALLY how the “real” system works, but being able to develop these parts together yet separately is the catch-22 aspect that I struggle with.

# SB Take 3

I’m pretty transient bar-ed out for now. I’ve done what I can with it, and I don’t have a good way to tackle the problems I’m facing with it. However, I DO think I’ve managed to create a small edge out of it, so I look forward to seeing how it does on my micro account. I think a partial martingale strategy may be in order, as eurusdd suggested, and while I don’t condone such tactics, I think they are a very advanced concept to implement correctly. Until then, I will wait for eurusdd to come back next year and talk about his version of transient waves, or if anyone asks/mentions anything interesting the thread. However with how secretive everyone is, I don’t see that happening.

Until then I’m back to pure metadata for another time! This time I have a more solid background, and a lot of ideas, however historically I’ve hit a wall with them quite quickly and it’s easy to become frustrated. I don’t know how long I’ll stick with it this time, but I will work until I’m out of ideas as usual.

I came up with a new approach after plenty of doodling on scratch paper, and these are the pre-lim results. It seems like 80%/20% extremes are decent at predicting turning points, while anything less than that, even though relatively extreme, does not do so much. I found some failures as well, but I will look into it later.

# Aggregating and fills Pt. 1

I haven’t been posting much (as I expected) but i’ve actually been getting a lot done. I’ve basically been running VBA macros non-stop for a few days now. The trick is that I need specific values based on the wave (and it’s length) so since every wave length is different (or unpredictable), I need the code to account for that. I’m looking at over 1500 swings and calculating 18 different values for every hour so excel has been quite busy. I’ve also had my fair share of debugging. Excel just finished the data and i’ve looked at it for a good 3 minutes and I already decided I’m going to need to write some more code. What use to take 5 minutes to analyze now will take 36 hours or so after fixing bugs (my friend thinks I should get to learn more coding and get a job in the industry. He has no clue how bad at it I am HA).

Upon looking at the data, I immediately thought of SB’s thoughts on fills and TCD trajectories in general: If done properly, one will be able to see the end of the ‘trend’ before it occurs. There will be signs that the subordinate TCD is growing in strength and prepared to overtake the dominant one.

# Attempt at states

I’ve been updating a paper to-do list which I think is helpful to give me a better, more clear direction on what to do next. Traditionally I’ve just asked questions about the market in general, and come back with a statistic that answers that question (what’s the average wick size?) What I’m trying to do now is to break the market down into a “complete” structure, by asking questions and receiving at the end not a statistic, but more data. Data that can be changed as the info that goes into it changes, and data that can allow me to ask more questions about the market.

I’ve been soaking in information from Rel and my first stop is to come up with my own way to do them. I think even though there are issues with the fixed TF, I just don’t have the programming ability, nor do I particularly want to expend the resources to learn how to create MTF or range bar charts. I’d also like to create these states as an indicator in MT4, but again, my programming skills are not up to par; When I have a specific goal in mind I’ll do the work to see if something is there first, and then perhaps pursue an indicator to make the grunt work easier should it be easier.

Right now, the hardest thing about creating an indicator (I think) is my foundation/premise. Rolling 1x24hr, my (temp) “fix” for MTF. There are a lot of issues with rolling days, specifically when it comes to H-L ranges and averages, because essentially rolling days means that for every additional hour, it’s calculated as an additional day. We’ll see if I can come up with a good way to filter out data and not destroy it or be too arbitrary.

I decided to start with 3 types of states: Flat, Expand, and Trend. I want flats to represent low volatility, expansions to represent increase in range with no direction, and trending to be the “move”.

the blue box is the SB flat box, but I want to make it more dynamic by doing the rolling hours and potentially creating a flat box that starts/stops at a specific hour, not day. Ideally a flat box would look something like the blue+yellow box combined. Atm, I kind of see expansion boxes like flat boxes with a bigger range.

Once I create the states, I need to extract information via asking it questions. I think flat and possibly expansion boxes can utilize the ‘funny rectangles’ and trends will utilize active price ratios as well as interacting TCDs. Identify state->utilize state specific strategy->determine next state timing. I now think I see the use in states. I think I will soon see the issues with it hehe.

Rough draft states. Currently I have a total of 94 state changes from January of this year until now. 1 state transition for every 36 hoursish? I know states are constantly interacting with each other and they overlap, but in this basic version, I’ve pinned down the state change to a specific hour. Problem using fixed time? Maybe. Workable? Hopefully. Perhaps I can late use some SB-style subordinate to dominate “state” trajectory” to determine probability of state change. I will definitely find out in time if the numbers are off.