Statistical Analysis of TZs

This thread contains the results of some research I've been doing with TZs (Transient Zones). What are TZs? They are zones of consecutive prices within a bar which aren't touched within a certain number of bars to the left and right of the TZ bar. They might also be called pivots, fractals, spikes, supply/demand areas, resistance/support areas, breakouts, and so on. But for the purposes of this thread let's call them TZs.

I'll assume that you already have an understanding of TZs and their nomenclature (h, k, transience, recurrence, etc.). If you don't then I recommend reading the start of this thread, which was kindly created by vlady1974.

The purpose of this thread is to prove the effectiveness, or ineffectiveness, of TZ systems in the most unequivocal way possible, using the statistical analysis of past data. Hence all probabilities shown are empirical, not theoretical. I encourage other people to post their results in a similar way. And I discourage people from posting opinions about TZs without having the figures to back them up. Personally Im not biased towards TZ systems either way. Im only interested in the edge they might provide. And if a particular system offers no discernable edge, then I wont be using it.

At the moment, I'm still undecided about whether to post any of my programs that produce these figures. There are many reasons for and against this. Currently I'm more in the against camp. It could be argued that posting results without the code is meaningless and untrustworthy, but I suppose that if anyone feels strongly enough about the results and the systems then they should write their own programs and/or analyse their own charts to prove or disprove them. That's what I would do anyway, rather than relying on someone else blindly.
Edit 04-May-2015: I've posted the main indicator TZJ3 (with viewable source code) to the thread in this post.

Unfortunately, like many of us here, my online time is very limited so please bear with me. There may be lengthy gaps between posts if something comes up. Also, if you have any questions or comments Id be grateful if you could ask them here rather than in PMs. Thanks.

Before we start, I'd like to thank FX-Jay, whose impressive trading results inspired me to start researching TZs in the first place. I've been working with him since January 2015 to produce the results I'm going to present, and I now realise more than ever what a top-notch trader he is. He’s also imaginative, smart, hard-working and professional. Thanks Jay, it’s been a pleasure

First Opposite TZs

The first system I’d like to present is First Opposite TZs. This isn’t the first system we worked on, but it is the main body of our work. It was inspired by an observation Jay had had about CTZ and PTZ triple sequences. CTZ (Confirmed TZ) is a TZ that is both h-left transient and h-right transient. PTZ (Potential TZ) is a TZ that is h-left transient but h-right recurrent.

This system is based solely on Tip TZs (bar wicks), and not Mid TZs.

The initial observation was simplified into double sequences, and we set out to prove the probabilities of two simple events:

1) First TZ being resolved following the occurrence of a second (opposite) TZ
PTZu > TZd and PTZd > TZu

2) Second TZ being resolved following a first (opposite) TZ
TZu > PTZd and TZd > PTZu

Note that: u = up (bar top); d = down (bar bottom); TZ = PTZ or CTZ (whichever); resolved = h-right transience is completely broken (price breaks the tip of the PTZ bar).

We assumed that the h value was the same for each TZ. Though in practice it could be more than the tested h value in one or both TZs, since an h=90 (for instance) also includes every h value down to 1.

We expected the probability of (1) to be relatively low, and the probability of (2) to be relatively high. In other words, when you have a PTZ on the chart for a certain h value, and an opposite PTZ appears for the same h value, then the probability of the second PTZ being resolved is much greater than the probability of the first PTZ being resolved. In other words, the second PTZ becomes the target (TP), and the first PTZ becomes the stop loss (SL).

Those were our expectations. It’s getting late now so I’ll post the results tomorrow hopefully.

You know, when you just read the name FX-Jay, you know for a fact that it's going to be golden.
Statistics > all.
Looking forward to seeing what you guys have figured out!

Much respect North!!!

Has been study this subject too on old eurusdd thread. it is gonna be a very interesting thread, subcribe

Please continue the topics, subscribed, thanks!

Thank you for your support everyone

So, just to recap, we are testing the theory that when you have a PTZ on the chart for a certain h value, and an opposite PTZ appears for the same h value, then the probability of the second PTZ being resolved is much greater than the probability of the first PTZ being resolved.

Here are the first set of results. Please take a few moments to study this file:



Just to explain briefly about the Alt 1 and Alt 2 figures. It occurred to me that double tops over a couple of bars or so were being counted as PTZs, not CTZs, which was distorting the figures slightly. After all, what we were really trying to find were pivots to which price didn't return within a certain number of bars. It didn't matter that the pivots weren't single-bar TZs.

So, "Alt 1" means "Alternative #1". PTZs that are double-tip over 2 bars (broken on the next bar because price reached the PTZ tip price but didn't exceed it) are counted as CTZs for the purpose of the stats. So in the chart below, the highest PTZu (yellow down triangle) that precedes the PTZd (yellow up triangle) is actually counted as a CTZu. So group PTZu > PTZd has one less than in the "Std" (Standard) count, and group CTZu > PTZd has one more.



And "Alt 2" means "Alternative #2". PTZs that are double-tip over 5 bars (broken within the next 4 bars because price reached the PTZ tip price but didn't exceed it) are counted as CTZs for the purpose of the stats. So in the chart below, the highest PTZu (yellow down triangle) that preceeds the PTZd (yellow up triangle) was JUST broken on the 2nd bar, so is counted as a CTZu. So group PTZu > PTZd has one less than in the "Std" and "Alt 1" counts, and group CTZu > PTZd has one more.

So, we have seen from the stats that there is indeed a very high probability of the first opposite TZ being resolved before it's "parent" (or base) TZ. Especially on M240, which is the timeframe we mostly focused on after that. And here is another file containing the same calculations for different currency pairs. These results support the idea too.



But, before we pop open the champagne and start dancing in the streets, can anyone see the flaw in these stats? I mean, not in the actual probabilities (which were extensively tested and are accurate as far as I know), but in what they are calculating?

Stuff to do, so I'll leave you with that thought for now

Hi North Trader,

I have been following old threads on TZ theory from sometime including threads on Penguin Traders . I tried to decipher the stats that you just mentioned and some of them closely resembled my results as well when i was trying to understand the concept ( but somehow never got to levels where i can trade based on those stats ). My inference(at that time) was probability might be in favor of us with second PTZ getting resolved but the potential payoff ( expected return ) from TP : SL is too skewed to give any identifiable edge in $ terms.

Very similar to old saying => you will make money most of the times but when you loose , you loose your shirts/pants

Please provide some more insight. Looking forward to understand and explore this concept once again.

I should be the one thanking you for all the hard work & dedication you've put into this project.
I'm not going to turn this into a praising contest you already know my opinion about you, and why I chose you to be one of the few to collaborate with.

All the best with the thread, let's see if someone will mange to catch onto something we missed

This is still a nice method, and i believe it can be profitable.

But imo what you need to analyze is the probability of resolving second ptz, in relation to how many bars has passed without resolving. Many of them are resolved withing the next 1-5 bars, and this creates the high %, but what is the succes rate of the 15th bar after second ptz without resolving?

I think this is the interesting analysis, because trust me, it wont be the same probability.

Good luck

Hello,

Great works Northtrader, and thanks very much for sharing.

Could you share the stats for gold?

Green pips to you

Hi, NorthTrader,

Your endeavors and investigations give me new breath of thinking on the subject. Thanks.

Very interesting.

Best,
D.P.

Absolutely, Ullen and sbc. You hit the nail on the head. Most of the PTZs were resolved very shortly after they appeared, which meant that price didn't retrace enough for them to be worth trading. In fact, the vast majority of them retraced less than 25%. Add to that the fact that your SL had to be all the way back at the Base TZ (preceding opposite) to guarantee those probabilities, then it doesn't look so rosy after all.

However, we knew all that. Those early figures were just a foundation on which to build. It was time to add some retracement parameters, and a bunch of other ideas, to try to turn it into a system that produced a edge. More on that next.

There are many ways to decide entry, fibo, s/r, wave and so on, but the interesting thing, is finding out if the probability decreases with the amount of bars it takes to resolve..

I'm sorry limprobable. I have no long term data for gold at the moment. If I get it, I'll run the final stats on it (which I haven't talked about yet) and let you know.

I can confirm that probability decreases as number of bars after the PTZ increases. To show an example of this, here are some stats from EU M240, 10 years of data:

CTZ h=35/35, PTZ h=35/35:
--- Total First Opposite PTZs = 220
--- Hit = 206 (94%) [1 was excluded due to > 100% retrace; higher than previously posted double sequence probability because the first zones are all CTZs]
Reduce PTZ h-right to 24 (66% of h-left):
--- Hit = 205 (93%) = 99.5% of h35 [all < 100% retrace]
Reduce PTZ h-right to 12 (33% of h-left):
--- Hit = 190 (86%) = 92% of h35

CTZ h=35/5, PTZ h=35/35:
--- Total First Opposite PTZs = 237
--- Hit = 216 (91%) [all < 100% retrace]
Reduce PTZ h-right to 24 (66% of h-left):
--- Hit = 212 (90%) = 98% of h35
Reduce PTZ h-right to 12 (33% of h-left):
--- Hit = 197 (83%) = 91% of h35

Conclusions
a. Most TPs are hit within 12 bars (if h=35)
b. Almost all TPs are hit within 24 bars (if h=35)
c. After 24 bars, waiting for a hit is futile. Better to exit at the best price you can.

However, I still think it's more to do with the price action than the number of bars passed. The more price retraces from TP, the less the probability becomes. I'll prove this later. But ultimately does it really matter which one is causing the drop in probability? Don't the two things (retracement and time passed) usually go hand in hand? What matters is that the problem exists, and we have to find a way to work round it.

Very interesting thread here. This concept is still very new to me but as some one who is very interested in statistical analysis it caught my eye. I have a few questions for some of the people who are familiar with what the OP is talking about.

1. The TZ zones are nothing more than Donchian Channels am I correct? h being the setting of the doncian channel? So we are calculating the probability of a break out of the previous high or low based on the specified time frame correct?

2. Since these events are time dependent the probabilities of each break out occurring will decay candle by candle drastically, with this how do you then know the true probability of the trade.(Example: If you enter the trade when 4 bars have passed, the probability will be x%, After 2 more bars form the probabilities will be drastically different then when you originally entered. How are you going to take this into account?)

3. Since the events are time dependent you need to factor in the probability of price finishing between the two levels when time expiries, which will also greatly affect the probability of the price hitting the stop or the target in the initial setup. Have you guys factored this in and I missed it?

The only way I can see these Donchian Channel Break Out statistics being useful is with a binary option where you do not need to worry about a stop loss or target. All you need to worry about is will price close above or below the last donchain channel touch? Just some thoughts. Will be watching to see what I can learn in here about this.

The stats are looking nice, and they always have been with TZ. The real problem though, is not wether price will resolve after 10 bars, it's if price will resolve once it has hit a certain level that i found interesting inbetween the two tz's. THAT is an interesting study!. I will be glad to continue the conversation in private, if you're not interested in sharing everything with everyone (spoon feeding). I'm not trading this concept, but it still has my interest.