Transient bars

A little more of a layman’s post this time around because it’s a new idea (:

-Transient: Noun-lasting only for a short time; impermanent.

-Interesting name for what they are. TIL a new word.

-Inspired by a section of the thread “Similarity System” over at FF by the beast eurusdd. It’s a series of ideas that doesn’t have much to do with the actual similarity system, so I’m glad I didn’t have to read 100s of pages to get caught up. I was lucky to catch it right when it began to unfold.

To me, I read about transient bars and immediately think “levels trading”. It’s quite similar to the B2i, and I hope to combine them and compare levels in the future.

The idea of trading transient bars has two parts: h and k. h is a number that refers to the number of bars to the left and right of a target bar. Although traditionally it’s always the same number on the left and the right, I suppose opening up possibilities to have a different number on the left than the right is completely possible.

A bar is considered transient(or h-transient) if there is no bar H bars away from the bar in question has a higher high or a lower low. Basically an extended fractal or swing identifier. Here’s the H=5 that I posted in the thread:

http://www.forexfactory.com/attachment.php?attachmentid=1490940&d=1408154242

If a bar is not H-transient, it is H-recurrent.

recurrent=price revisted. Transient=not revisited.

Of course, you don’t know a transient bar has formed until it’s actually completed, just like fractals. Had I been told about this and never read the thread I would have tried an array of things: timing between T-bar to T-bar, range, frequency, time of day, etc. But this thread opened me up to trying something a bit different.

A few things should be known about the market before proceeding that were given by eurusdd. First that most bars are H-recurrent. It’s kind of funny to think that everyone is scared to buy at the top or short at the bottom when really, bars like that are extremely rare! It’s really being proven to me that trends have strength and don’t give up so easily.

Logically, since you can’t trade transient bars (in the future) you have to trade what I call LTB: Left-side Transient Bars.

First, some stats:

eurusdd said that if your H is correct, 97% of the time a bar should be recurrent. I played with the numbers and have this for my current H on my TF

Overlapping bar is an extra stat and refers to bars where a transient bar occurs right after a transient bar has just occurred. Very rare.

So, if true transient bars only occur about 3/100, then you should trade any bar that’s left side transient in the current direction since it’s unlikely to become an actual transient bar right?

Perhaps not. Filtering only for bars that are LTB, the rate increases to about 13%. So if a bar is already an LTB, it’s success rate to be pushed further is 87%. Still not bad but down a bit from 97%.

Normally I think most people would jump on that and try to make something of it. Heck, maybe there IS something to it. But eurusdd suggested that 87% is meh. Strive for more! Strive for.. the possibility of near 100%??

A ‘grail’ is hidden here. There were at least 10 posts of people questioning how it is possible to win in the scenario that you can sell in a top-transient bar and still win.

If a possible transient zone is forming on top (up scenario), mark it as such and mark the area above it as recurrent and sell? Not intuitive. Yet the charts and back testing so far say other wise.

The gray arrow marks the left side transient bar. according to eurusdd, mark up the area as transient, and the area below it (in down scenario) and look for buys. And here, we get it and get out.

How big to draw the entry area? How safe is it? How long do you wait? Remember the second part of this whole thing? k? I think the secret lies in how one deals with the k variable. The H part of it is not too difficult when you understand the concept, but the k is much more dynamic. And as with everything else I’ve learned thus far, my guess is that it should be dynamic, safe, and consistent. Looking for frequency distributions most likely.

Back to work.

10 thoughts on “Transient bars”

1. 4xBones

LG, how do you analyse your data? Spreadsheet or something else? I’d like to try some number crunching myself.
Thx,

Yup! Everything is done using excel. The basic things can be done with if-statements and filters. For the more dynamic parts I write macros

2. anon

Just for completeness, what H were you using and what period of historical bars did you derive your statistics from?

The h value will be my secret. Not because it works,but to encourage you to discover your own h value and statistics. Don’t worry! I believe that should you find an h value that meets the 97/3 rule (not difficult with the recurrence statistic indicator in the thread), they will more or less say the same story as a variety of other h values.

As for historical data, I’m using H1 for about 3600 days dating back to ~1999

1. anon

Ok, I understand. I have already run a lot of stats and my suggestion to you would be to move away from bars and try ticks directly. I have been working with ticks over the period of 2010 to 2014 (a 4gb file!) and I find it much easier to generate the stats. For me, I simply iterate over a loop over h values (in seconds) and a k value in pips (eg h could be 312mins with a k of 7 pips) and generate stats for every possible h and k value. For sure it takes along time to generate all the stats but I think it generates a more complete set of stats as bars a simply a sub-sampled collection of pips. Now that I have all the stats I still find the same issues as you though, eurusdd mentioned that dna is also required (which I could never model correctly) before the next revolution step could be calculated, so I guess that we are simply waiting for more pieces of the puzzle.

Another interesting thing to try when you are working with ticks is to filter out spikes, I generate stats ignoring data that moves too fast or slow, eg if my k is 4 pips, I ignore any time period there that k was reached in under x seconds. I’m doing this to try to filter out fast news spikes. The stats look better but the trading issues still remain.

Anyway, it was good to find your site, (and it’s interesting to see who else is lurking on here). I’m especially thankful that you introduced me to iDouble (and TradeVector etc), there are some interesting posts that I had not seen before and have my interest. I think Patrick would also be interested in them also….

Thanks for the info anon. I use 1hr data because it’s what I have access too. I have never used tick data and it’s a pretty foreign concept to me, I have never really tried to use it because I heard it was difficult to get a hold of quality tick data. Isn’t tick data similar to range bars in the sense that they don’t generate based on time but rather price movement? That seems a bit confusing to apply the time concept with pure price movement, I’m not sure how to go about it.

Perhaps it’s also in my coding ability that I do everything in excel that I stay away from tick data. I did a google search to try to download some tick data and excel couldn’t even fully open 1 month of data before it reached it’s 1,048,576 row limit. Talk about a lot of data!

Its good chatting with you, this blog is isn’t advertised so I’m curious as to how people find my stuff. (:

1. anon

For a good source of tick data (or any fx data) I use Dukascopy. You can open a demo account and from within their platform you can download data from any time period for any instrument directly into csv. From that I use c# to process the data.

Anyway, the concept of time is the same with ticks in this concept. You are simply looking to see if price has touched at x secs/mins/hours before or after the current price (+/- the pip offset).

If you are really keen to do statistical analysis I suggest you start learning c# or R or something more powerful then excel. Excel is good but at some point you’re going to need more horsepower!

I think I found your blog googling around for “transient zones” or something along those lines.

2. Patrick M. White

I think I did about the same web search that anon did. But now that Rel has posted links to this blog in her FF lab, the secret will soon be out!

@anon I’ve been lurking on the B2i thread for some time but thanks for thinking of me! I’m also in Rel’s Lab but don’t tell her. 😉 Sometimes I don’t post in interesting threads even though I’m following them closely.

It is very nice having this quiet, semi-private space away from the madding crowd where we can talk about concepts and the things we are learning from the data. LG is doing some excellent research and it is inspiring me to look more closely at the data. One thing I’ve been looking at is studying hit stats based on left – h and right – h (two different values of h for before the transient bar / after the transient bar). I think it’s worth looking at closely.

3. Relativity

I agree with the use of R, but I do think it also depends on how good the modelling of data is. I’ve survived and thrived well just using MT4 and Excel, although I do know R. If the modelling is robust, Excel should suffice when using nothing but Frequency Distributions and Plots.

However if you really want to go in depth to tick level data, I will not disagree R (backed with a strong CPU + database) is necessary. I stopped using R when I found that its possible to model the data and reduce it to what I needed most. For my case, swing point to swing point. I didn’t need to care too much about the ticks that happen in between the points; I wanted the start and end points -> entries and exits.