The similarity system | Page 576 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Similarity: A Practical Approach 859 replies

Similarity Trading - The ZZBB Method 215 replies

Rainbow similarity 256 replies

Similarity Stuff 17 replies

Directional Similarity 0 replies

Attachments: The similarity system
Exit Attachments

The similarity system

  • #11,501
  • Nov 3, 2014 1:43pm Nov 3, 2014 1:43pm
Quoting saver0
{quote} The 9% chance of TZ is when it was being compared to all the number of bars which it looked at. It wasn't comparing it to the number of Recurrent zones. I think this is what most may not have gotten (or what I don't get). For a zone to be recurrent, it has to be H-Transient on the previous number of H bars, correct? Then the number of RZ bars won't be equal to Number_Of_Bars_In_Chart - TZ_Bars. We would have to actually have to find RZs that showed up exactly at the previous TZ levels and this number is very small. Thats why there is 85%...
I understand what's the reason behind the confusion, there are 2 theories being presented and thus the mix-up in terms & numbers.

For now let's focus on the idea behind the Daily GU chart that you first quoted.
I'll use your terms to try to clarify:
"all the number of bars which it looked at" = "the number of Recurrent zones"
Think of it, a price can be either recurrent or transient, if it is not Transient (9%) then it is Recurrent.
This is different than the idea of Potential Transient Zones!

"We would have to actually have to find RZs that showed up exactly at the previous TZ levels and this number is very small"
That's true, but the number isn't as small as you think. However, the aim is to get high probability trades no matter if it's 1 per hour or 1 per week!

Personally if I want to back-test, I would rather keep K on the side and focus on increasing H
  • #11,502
  • Edited 2:05pm Nov 3, 2014 1:46pm | Edited 2:05pm
Quoting DavidRP
{quote} I have those numbers and they will surprise you, double tops and triple tops have a role here, it's not so rare to have another tz (for the same price) after having the first one. (Double tops i.e)
Is it possible to share actual numbers and percentages? (for different Hs, pairs, and TFs)

Clarification: When we're talking about 2 or 3 TZs for the same price, we're talking about consecutive TZs, so price shouldn't be recurrent at that exact price before it gets transient again.
  • #11,503
  • Nov 3, 2014 2:05pm Nov 3, 2014 2:05pm
Quoting DavidRP
{quote} I have those numbers and they will surprise you, double tops and triple tops have a role here, it's not so rare to have another tz (for the same price) after having the first one. (Double tops i.e)
Surprise in terms of good or bad?

What kind of numbers are you getting?
Focus, Patience, Determination & Order in chaos
  • #11,504
  • Nov 3, 2014 2:06pm Nov 3, 2014 2:06pm
I will advise people to be careful with the notions of transient/recurrent vs the notions of h-transient/h-recurrent.
transient/recurrent applies to the entire space under consideration. It is a global idea like global minimum/maximum. This global truth is the reason most traders lose money. Why? They know that certain prices will be re-visited but they just do not know how to quantify the waiting-time.
h-transient/h-recurrent is the localization of the global principle.
That is we can restrict ourselves to an open set within our domain of interest and localize the idea of a recurrent states, with a high degree of certainty, close to 1, depending on the value of h we choose.

So before you go around shouting recurrent/transient, please be careful and add the appropriate h-. If you ignore this, you will create confusion!
  • #11,505
  • Nov 3, 2014 2:08pm Nov 3, 2014 2:08pm
Quoting FX-Jay
I'll use your terms to try to clarify: "all the number of bars which it looked at" = "the number of Recurrent zones" Think of it, a price can be either recurrent or transient, if it is not Transient (9%) then it is Recurrent.
Yea, recurrent but it wouldn't be H-Recurrent. We are only interested in H-Recurrent states, correct?
Focus, Patience, Determination & Order in chaos
  • #11,506
  • Nov 3, 2014 2:11pm Nov 3, 2014 2:11pm
Quoting saver0
{quote} Surprise in terms of good or bad? What kind of numbers are you getting?
He's implying that it is not rare to have 2 or 3 consecutive TZs. (bad surprise)
In my opinion, he is mistaken unless he is able to prove it with facts.

I can show you many combinations of pairs/H values/TFs where the frequency of 2 consecutive TZs is 0 or 1. ( good surprise )
  • #11,507
  • Nov 3, 2014 2:14pm Nov 3, 2014 2:14pm
Quoting DavidRP
{quote} I have those numbers and they will surprise you, double tops and triple tops have a role here, it's not so rare to have another tz (for the same price) after having the first one. (Double tops i.e)
DavidRP: I think you are missing an important point or the idea of the theorem.

That double-tops or triple top do not invalidate the theorem. Let me show you why.

Mark the set of prices where the double-top or triple tops occurred. Now, go back into the history and check every time the market got to those prices. calculate the relative frequency of h-recurrent.
Now this relative frequency should go to 1, almost surely, as time goes to infinity.

So, just because that zone was h-transient 3 times in a row is not enough reason to form a firing squad!
This theory is independent of the underlying forces and you cannot disprove it with a few sample points!
  • #11,508
  • Nov 3, 2014 2:19pm Nov 3, 2014 2:19pm
Quoting FX-Jay
{quote} Is it possible to share actual numbers and percentages? (for different Hs, pairs, and TFs) Clarification: When we're talking about 2 or 3 TZs for the same price, we're talking about consecutive TZs, so price shouldn't be recurrent at that exact price before it gets transient again.
of course I'm speaking about consecutive prices TZs Look any chart, and take a look of double tops you can find there consecutive Tzs
  • #11,509
  • Nov 3, 2014 2:21pm Nov 3, 2014 2:21pm
Quoting Eurusdd
{quote} DavidRP: I think you are missing an important point or the idea of the theorem. That double-tops or triple top do not invalidate the theorem. Let me show you why. Mark the set of prices where the double-top or triple tops occurred. Now, go back into the history and check every time the market got to those prices. calculate the relative frequency of h-recurrent. Now this relative frequency should go to 1, almost surely, as time goes to infinity. So, just because that zone was h-transient 3 times in a row is not enough reason to form a firing...
I'm not disproving nothing, it was a surprise for me, and when taking a look close I discovered that it was because double and triple tops

I will run the test later and post the results from 2003 with 5 min bars
  • #11,510
  • Nov 3, 2014 2:28pm Nov 3, 2014 2:28pm
Attached Image (click to enlarge)
Click to Enlarge Name: Capture (1).PNG Size: 32 KB

Attached Image (click to enlarge)
Click to Enlarge Name: Capture.PNG Size: 32 KB
  • #11,511
  • Nov 3, 2014 2:30pm Nov 3, 2014 2:30pm
Quoting DavidRP
{quote} of course I'm speaking about consecutive prices TZs Look any chart, and take a look of double tops you can find there consecutive Tzs
Of course they exist, you can even find 4 and 5 TZs in a row for certain H values.

I was hoping that you would surprise me with concrete statistics.
  • #11,512
  • Nov 3, 2014 2:32pm Nov 3, 2014 2:32pm
Quoting FX-Jay
{quote} Of course they exist, you can even find 4 and 5 TZs in a row for certain H values. I was hoping that you would surprise me with concrete statistics.


results were something like 15% or so (after one TZ) I will post them later
  • #11,513
  • Nov 3, 2014 2:35pm Nov 3, 2014 2:35pm
Quoting Eurusdd
I will advise people to be careful with the notions of transient/recurrent vs the notions of h-transient/h-recurrent. transient/recurrent applies to the entire space under consideration. It is a global idea like global minimum/maximum. This global truth is the reason most traders lose money. Why? They know that certain prices will be re-visited but they just do not know how to quantify the waiting-time. h-transient/h-recurrent is the localization of the global principle. That is we can restrict ourselves to an open set within our domain of interest...
It means we should only consider a trade set up if only price visit a previously TZ zone and made that zone PTZ. If our h and k values are right we will have a high probability trade for that zone. But it is our best interest if we let "k" part of that zone to remain transient and take only the remaining part as our trade target.
  • #11,514
  • Nov 3, 2014 2:40pm Nov 3, 2014 2:40pm
Here's something for the average Joe that doesn't have enough statistical/mathematical knowledge.

Word of cation: This is for demonstration purposes only, and isn't scientifically proven! + It doesn't express my opinion towards the probability and TZ concept.

Let's assume you have a bag of 1,000 underwear. 970 are women red thongs & 30 are white men's briefs.
If you randomly pick 3 underwear from the bag and they turned out to be all for men, would you bet that the 4th pick will be a thong or a brief??

I would bet for a red thong, not because I like them, but because I believe that the probability is higher for a thong to show up.
Does that mean that I can't pick 4 briefs in a row? NO - I can pick 30 in a row and that's just sad! But it could happen.

Betting on high probability doesn't mean you will win 100% of the times like everyone dreams.

It's your choice to be a stupid gambler, a smart gambler, or a casino. Btw guess what the casino doesn't win 100% of the times, there are always 3% of the gamblers that manage to make money

P.S. Kiads please feel free to provide visual support to the above example
  • #11,515
  • Nov 3, 2014 2:43pm Nov 3, 2014 2:43pm
Quoting FX-Jay
Here's something for the average Joe that doesn't have enough statistical/mathematical knowledge. Word of cation: This is for demonstration purposes only, and isn't scientifically proven! + It doesn't express my opinion towards the probability and TZ concept. Let's assume you have a bag of 1,000 underwear. 970 are women red thongs & 30 are white men's briefs. If you randomly pick 3 underwear from the bag and they turned out to be all for men, would you bet that the 4th pick will be a thong or a brief?? I would bet for a red thong, not because...
you funny
  • #11,516
  • Nov 3, 2014 2:44pm Nov 3, 2014 2:44pm
Quoting FX-Jay
I would bet for a red thong, not because I like them, but because I believe that the probability is higher for a thong to show up. Betting on high probability doesn't mean you will win 100% of the times like everyone dreams. It's your choice to be a stupid gambler, a smart gambler, or a casino. Btw guess what the casino doesn't win 100% of the times, there are always 3% of the gamblers that manage to make money P.S. Kiads please feel free to provide visual...
One smart Homo-Sapien!
You belong to the Real-Trader's CLUB!
  • #11,517
  • Nov 3, 2014 2:45pm Nov 3, 2014 2:45pm
Quoting DavidRP
{quote} I'm not disproving nothing, it was a surprise for me, and when taking a look close I discovered that it was because double and triple tops I will run the test later and post the results from 2003 with 5 min bars
Yes double, triple tops and bottoms are easily eyeballed when curve-fitting (H) on your chart; a "post/after-approach". Therefore that should not be your goal. Instead, analyze behaviour and calculate probabilities on a possible (2), (3), or even (4) in a row. While this looks simple the recurrent area size(K) actually also has its OWN behaviour, corresponding to the rareness of the event. What i personally found was that the higher the number gets (especially 3-4 (havent seen 5 yet), the smaller your (K) will be in general(there are also other factors at play here, such as similarity that influence (K). Now I ran this sample over only one year, and one pair(eurusd), so the results are still insufficient but so far it looks promising. If its set-up correctly it should work for every pair/market, unless pair DNA also plays a part.

Whether there is a correlation between pair DNA and (K), is a piece of the puzzle i would still like to verify myself. Statements are made here but one needs to see it him/herself.
  • #11,518
  • Nov 3, 2014 2:45pm Nov 3, 2014 2:45pm
Quoting DavidRP
{quote} results were something like 15% or so (after one TZ) I will post them later
This goes in hand with savor0's calculations.

Again, 85% is not 90% or 99%, but it is still something worth betting on, no?

My account says yes eventhough I had in my head a number higher than 85%

While doing your testing, please include the statistics for EU H4 H=36

Thanks in advance.

Best,
Jay
  • #11,519
  • Nov 3, 2014 2:53pm Nov 3, 2014 2:53pm
This is another way to understand the theorem..
For any price p, there is a minimum value for h such that the relative frequency of h-recurrent states associated with p goes to 1, almost surely, as time goes to infinity. Therefore, even if the relative frequency is around 85% currently or even 50%, it does not change the ultimate probability.
Now, if in the long run, the relative frequency does not go to zero, then the theorem says, that price should belong to a set of prices with that same feature and that set should be of measure zero!!!

Now, notice that a price is a real number, NOT A ZONE!
  • #11,520
  • Edited 3:48pm Nov 3, 2014 2:59pm | Edited 3:48pm
Quoting FX-Jay
{quote} This goes in hand with savor0's calculations. Again, 85% is not 90% or 99%, but it is still something worth betting on, no? My account says yes eventhough I had in my head a number higher than 85% While doing your testing, please include the statistics for EU H4 H=36 Thanks in advance. Best, Jay

Here the stats (5min from 2003 data from Dukascopy) with colums: h, totalPrices,totalTransientPrices(%), total2consecutiveTprices, % of a price being T after being T,total3consecutives, % of begin T after being TT,total4consecutives, % of begin T after being TTT.

For the case of h=6 we have 2.88% of transient prices, then 41392 prices TT with a probability of 20.93% of being T if it was T the last time. Indeed the chance of getting a T if it was T seeral times in a row is higher-> Support and resistence. Makes sense to me.


I double checked the results and seem ok

h,totalPrices,totalTransients,,Tp(%) ,2 consecutive, 3 consecutive, 4 consecutive
1 6863714 529414 7.71 164399 31.05 54161 32.94 19245 35.53
2 6863708 388061 5.65 106297 27.39 31240 29.39 10083 32.28
3 6863701 308973 4.5 77107 24.96 20763 26.93 6088 29.32
4 6863694 258801 3.77 60352 23.32 15240 25.25 4250 27.89
5 6863690 224146 3.27 49547 22.1 11892 24 3167 26.63
6 6863686 197755 2.88 41392 20.93 9384 22.67 2338 24.91
7 6863678 176614 2.57 35052 19.85 7464 21.29 1720 23.04
8 6863670 160498 2.34 30620 19.08 6257 20.43 1390 22.22
9 6863662 147537 2.15 27402 18.57 5463 19.94 1185 21.69
10 6863654 136753 1.99 24786 18.12 4856 19.59 1013 20.86
11 6863648 127106 1.85 22430 17.65 4240 18.9 891 21.01
12 6863639 119254 1.74 20585 17.26 3835 18.63 800 20.86
13 6863633 112672 1.64 19105 16.96 3522 18.43 726 20.61
14 6863627 106731 1.56 17718 16.6 3198 18.05 650 20.33
15 6863622 101322 1.48 16518 16.3 2918 17.67 586 20.08
16 6863613 96474 1.41 15493 16.06 2685 17.33 531 19.78
17 6863607 92018 1.34 14462 15.72 2443 16.89 478 19.57
18 6863601 88076 1.28 13562 15.4 2239 16.51 432 19.29
19 6863595 84725 1.23 12895 15.22 2102 16.3 401 19.08
20 6863590 81784 1.19 12312 15.05 1993 16.19 383 19.22
21 6863584 78795 1.15 11719 14.87 1886 16.09 356 18.88
22 6863580 76131 1.11 11190 14.7 1767 15.79 328 18.56
23 6863572 73774 1.07 10708 14.51 1661 15.51 297 17.88
24 6863561 71636 1.04 10315 14.4 1598 15.49 287 17.96
25 6863554 69715 1.02 9961 14.29 1522 15.28 274 18
26 6863540 68068 0.99 9624 14.14 1465 15.22 260 17.75
27 6863534 66425 0.97 9318 14.03 1424 15.28 254 17.84
28 6863528 64908 0.95 9032 13.92 1365 15.11 244 17.88
29 6863522 63283 0.92 8693 13.74 1297 14.92 237 18.27
30 6863515 61729 0.9 8339 13.51 1209 14.5 205 16.96
31 6863508 60361 0.88 8096 13.41 1168 14.43 199 17.04
32 6863496 58986 0.86 7868 13.34 1120 14.23 188 16.79
33 6863488 57696 0.84 7644 13.25 1072 14.02 179 16.7
34 6863481 56559 0.82 7449 13.17 1044 14.02 178 17.05
35 6863473 55476 0.81 7294 13.15 1021 14 169 16.55
36 6863462 54445 0.79 7149 13.13 1007 14.09 168 16.68
37 6863454 53416 0.78 6961 13.03 972 13.96 164 16.87
38 6863439 52577 0.77 6844 13.02 944 13.79 160 16.95
39 6863433 51584 0.75 6655 12.9 909 13.66 151 16.61
40 6863427 50596 0.74 6485 12.82 878 13.54 146 16.63
2 traders viewing now