S/R Area Algoritm | Forex Factory

Options

Permalinks
Print Thread

Similar Threads

Statistically testing every indicator with my Genetic Algoritm 132 replies

The area of comfort. 2 replies

how to place many stoch in one area 1 reply

Attachments: S/R Area Algoritm
Exit Attachments

S/R Area Algoritm

  • #1
  • First Post: Aug 8, 2011 6:21am Aug 8, 2011 6:21am
Hi,

I'd like to code an indicator/ea which can find current important support and resistance area's, like in the added picture. I'm getting stuck with the logics however. Does someone know an algoritm or method to mechanically find these area's?

Untill now i only tried to put all fractals in an array and form clusters of price area's out of that array. The main problem here however is that fractals also appear at not-so-important area's, confusing the logics.

Hope anybody can help me out here?

Thanks in advance!

Simon
Attached Image (click to enlarge)
Click to Enlarge Name: sandr.png Size: 18 KB
  • #2
  • Aug 8, 2011 7:28am Aug 8, 2011 7:28am
Hi Simon ,

I have previously thought about doing this and my logic at that stage was :

1) Define a period that you want to look back at - i.e. 100 Bars
2) Start with open and closing prices and do a count for each specific area ( 0.0001 - 0.0005 pip difference depending on your view of a s/r area)
3) Incorporate High and low prices using the same method
4) Over the period you will have certain areas with a much higher count than others. This will be your S/R levels.
5) If for example a certain price area was touched 7 times in the last 100 bars you will have a very strong s/r point while a place that is touched 1 will be of no importance.

Hope this helps you a little bit.
  • #3
  • Aug 8, 2011 9:05am Aug 8, 2011 9:05am
Thanks for your reply. Seems like a good method, there is a major pitfall i can think of however:

What if price stalls in a certain horizontal channel of a couple of hunderd of pips, taking 10 bars to go up and 10 bars to go down. Your algoritm will then threat the whole area as one big s/r, while actually the top and bottom lines are the one youre interested in.

I did gave it another thought this afternoon and think i'll try out this idea:

- take every enfugling pattern (don't know if this is the correct name, the pattern where an up bar is followd by a down bar or vica versa)
- rate every individual pattern by the degree of where price came from and where price went afterwards (did it go away 20 pips, 100 pips perhaps)
- After atleast 2 patterns at more or less the same price with atleast x points there is an s/r

Thanks however, maybe someone else with an inspiring idea?
  • #4
  • Nov 11, 2011 6:48am Nov 11, 2011 6:48am
Quoting esquire
Thanks for your reply. Seems like a good method, there is a major pitfall i can think of however:

What if price stalls in a certain horizontal channel of a couple of hunderd of pips, taking 10 bars to go up and 10 bars to go down. Your algoritm will then threat the whole area as one big s/r, while actually the top and bottom lines are the one youre interested in.

I did gave it another thought this afternoon and think i'll try out this idea:

- take every enfugling pattern (don't know if this is the correct name, the pattern where an up bar is...
Hi esquire,

I'm also interested in this. I strongly believe that an S/R is a zone, not a line. Market is moved by people. People just consider 0.997 and 1 are just the same. That's why I'd say that the pitfall you describe may probably be a strength!

To me an S/R is an area the price finds hard to walk through or just can't cross. If a doji or a spinner top candle sits on such a line, I would consider this as a confirmation. The price couldn't move fast through this line. Same if there are some candles of alternate colors: up, down, up, down then breakout from the zone. Again a difficult journey. I suggest to search the S/R in an higher time frame than the one you target, but I'm unsure of this point.

I wouldn't limit the scope of search. Look USD/JPY, weekly. The S/R on 80.00 was first touched in 1995. As another example, price 101.00 was touched in 1994, 1996, 2000, 2005 and 2009... I'd keep all I have, no limit.

Ok, now here is my idea. It's a bit complex and I've no clue if it is easy to turn into an MT4 AE. Personally I target java any way ;-)

For each "price of interest" (like a swing high/low point) I define a gaussian function centered on this value. This way I blur the line. I leave the variance as a parameter, it represents the distance between two prices you consider the same. Then I sum up all of the gaussian functions (hundreds!). I attached an example of such a function.

You only consider the values above a given threshold, representing the number of times the line got hit. This threshold must be lower because of the blurring effect. Say I use 2.7 for 3 touches. The lines are the local maxima of the function.

I tried manually on EUR/USD, October 2011, H4. I use the 38 fractals I found. I got 2 levels (1.3823 and 1.3956). Certainly because the sample is very small. I also attached a snapshot of H1 November to see how the lines behave in the future
Attached Image(s) (click to enlarge)
Click to Enlarge Name: oct_h4.png Size: 31 KB
Click to Enlarge Name: nov_h1.png Size: 32 KB
Attached Image
No greed. No fear. Just maths.
  • #5
  • Nov 11, 2011 7:28am Nov 11, 2011 7:28am
There is an indicator called supdem, do a search for it.

If you do manage to code this could you please share it, I've been looking for something like this for ages.

The problem with supdem was it repaints and gets it wrong a lot but it was definitely on the right track.
{Promotion Removed}
  • #6
  • Nov 11, 2011 11:18am Nov 11, 2011 11:18am
Quoting leerees
There is an indicator called supdem, do a search for it.
i had a look and must say I wasn't impressed...

Quoting leerees
If you do manage to code this could you please share it, I've been looking for something like this for ages.
I checked my results and have to say that I missed two other S/R lines. There are 4 prices: 1.3373, 1.3682, 1.3823 and 1.3956. You may add 1.3893 if you lower the thresold from 2.7 to 2.5

Drawing the line at 1.3373 I saw that it got hit twice on the 22nd and 26th of September. That is before the October fractal swing points I used. Is it an evidence the algorithm makes sense? Probably worth I start working on coding it...

Quoting leerees
The problem with supdem was it repaints and gets it wrong a lot but it was definitely on the right track.
I don't understand what you mean by "repaint". If you mean rewriting the line (moving them) when new data come, this algo will do too.
No greed. No fear. Just maths.
  • #7
  • Nov 11, 2011 12:19pm Nov 11, 2011 12:19pm
leerees just informed me about the replies in this topic. First of all, nice work PipMeUp! Have to say I dont know a lot about Gaussian functions or that i'm a very good programmer - your results already beat me - but your stuff looks really impressive!

The solution I came up with quite some time ago was relatively simple: All the lines are very easy to spot with the eye, but extremely hard to generate with code. Besides that, most lines almost never change - most lines i studied are usefull over several months (H1 timeframe). So all that needs to be done is drawing the lines yourself, and then write a simple indi which stores all the values.

Seemed like the easiest solution at that time. Got stuck after that, not being able to decide a profitable strategy with the information of the lines..
  • #8
  • Nov 11, 2011 1:03pm Nov 11, 2011 1:03pm
I think here is a good code

http://breakoutfutures.com/Newslette...letter0303.htm
  • #9
  • Edited 3:55pm Nov 11, 2011 3:43pm | Edited 3:55pm
Quoting PipMeUp
Hi esquire,


I'm also interested in this. I strongly believe that an S/R is a zone, not a line. Market is moved by people. People just consider 0.997 and 1 are just the same. That's why I'd say that the pitfall you describe may probably be a strength!

To me an S/R is an area the price finds hard to walk through or just can't cross. If a doji or a spinner top candle sits on such a line, I would consider this as a confirmation. The price couldn't move fast through this line. Same if there are some candles of alternate colors: up, down, up, down then...
Hello PipMeUp,

This has already been tried recently. The results were good but the fact is recent SR's carry more weight. The SR's are an exacting thing and not a large range as you suggest. The recent range is very small as limit orders are placed just before an SR line. There will be over shoots of the SR line in times of high volume but the PA will promptly reverse and then retest the SR line most of the time. This is also due to differences of trader opinions. As the majority of big money closes or reverses their trades the rest realize their mistake follow their lead.

The value of an indicator your suggesting is when no recent SR lines exist. As the PA approaches a previously strong SR line it will reverse direction and any further PA hits will honor the new line to an exactness and disregard the old SR range line. People do move markets but it's not the helter skelter way most might think. Big money follows Support and Resistance.

Very old SR's are worthless because as time passes currency values change. Just because the GBPUSD was at $3.00 in 1970 doesn't mean it will ever see that level again. Even if it does reach that level it has no correlation to the past.

Have a great day,

Johnny
  • #10
  • Nov 11, 2011 3:56pm Nov 11, 2011 3:56pm
Johnny that's not entirely true. I agree that recent S/R zones carry more weight. I've seen V tops and bottoms leave a lasting S/R line for years.
I think it's the sharpness and depth of the move that produces the lasting S/R level.
You are in a maze of twisty little passages, all alike.
  • #11
  • Nov 11, 2011 9:54pm Nov 11, 2011 9:54pm
Quoting stevegee58
Johnny that's not entirely true. I agree that recent S/R zones carry more weight. I've seen V tops and bottoms leave a lasting S/R line for years.
I think it's the sharpness and depth of the move that produces the lasting S/R level.
Ok, show an example of a chart.

Johnny
  • #12
  • Nov 13, 2011 9:42am Nov 13, 2011 9:42am
Hi,

I discovered FX this summer, I'm only demo trading. I definitely don't consider myself any expert :-)
That was just an idea I threw in the air. I didn't expect a lot of answers. If any.

As you are a live trader, I'm extremely interested in knowing/understanding how you do find these levels. And how you avoid your chart cluttered by useless lines.

Quoting JohnnyBSmart
This has already been tried recently.
Please point me where.

Quoting JohnnyBSmart
- The results were good but the fact is recent SR's carry more weight.
- Very old SR's are worthless because as time passes currency values change. Just because the GBPUSD was at $3.00 in 1970 doesn't mean it will ever see that level again. Even if it does reach that level it has no correlation to the past.
$1 in the 50's you could get a meal. $1 nowadays you can't get the coffee. Very valid point. Taken. Unfortunately this brings the bad question how recent is recent?
Yet this doesn't anhilates the idea, bell curves can be weighted by the age of the points. This weight can become 0 after some time. How to fade them remains the big question.

Quoting JohnnyBSmart
- The SR's are an exacting thing and not a large range as you suggest.
I don't remember having suggested a large range. Is there a misunderstanding because two lines are drawn on the chart? They were supposed to be the center of two S/R areas. How wide are they? No clue!
Again that brings the how large is large? question. Are 10 pips big for you? To me that's small because it's in the order of magnitude of the spread. Just my point of view. M1 scalpers may see 10 pips somewhat different.

For what I'm sure, a S/R just cannot be a 0-width line. Impossible.
1/ Institutions aren't fed the exact same data. Two lows for two big traders are probably not at the exact same price.
2/ Big traders have near 0 spread but medium ones do have spead, so the same stop entry at the very same place would not enter the market at the very same price for both of them.
3/ More important, you write yourself "The recent range is very small as limit orders are placed just before an SR line." Therefore not on the line. This is enough to turn it into an area. And this brings the same stupid kind of question What is just below?
That's the same idea for trend lines. I just can't figure out why some traders/posters insist on drawing these lines at the very exact perfect pip on a daily chart. They allow no tolerence on the Y-axis (price) while they seem not concerned by the fact they allow for 86400 seconds error on the X-axis (time) !

It must also depends on what you want to do with these lines.
No greed. No fear. Just maths.
  • #13
  • Nov 13, 2011 3:12pm Nov 13, 2011 3:12pm
Hello PipMeUp,

I will try to answer your questions.

Quote
As you are a live trader, I'm extremely interested in knowing/understanding how you do find these levels. And how you avoid your chart cluttered by useless lines.

I don't have useless lines on my charts. I only have recent (30 days) S/R lines and 100-200 day hot lines of previous S/R.

Quote
Please point me where.

Try Hanover's thread. Development is on going on another version.

Quote
$1 in the 50's you could get a meal. $1 nowadays you can't get the coffee. Very valid point. Taken. Unfortunately this brings the bad question how recent is recent?
Yet this doesn't anhilates the idea, bell curves can be weighted by the age of the points. This weight can become 0 after some time. How to fade them remains the big question.

Good question. It depends on the chart. If there are recent (hours or days)SR lines the PA will honor these. If the PA is moving into old territory of say 6-12 months ago it will honor those. Remember traders are moving the PA to a certain level and they will use the most recent SR's available even if that means some months ago.

Quote
I don't remember having suggested a large range. Is there a misunderstanding because two lines are drawn on the chart? They were supposed to be the center of two S/R areas. How wide are they? No clue!
Again that brings the how large is large? question. Are 10 pips big for you? To me that's small because it's in the order of magnitude of the spread. Just my point of view. M1 scalpers may see 10 pips somewhat different.

My findings are if your old data draws a line and it is 10 pips from the most recent then it might as well be a 100 as the PA will either blow right by it or not hit it if a more recent SR is available.

Quote
For what I'm sure, a S/R just cannot be a 0-width line. Impossible.
1/ Institutions aren't fed the exact same data. Two lows for two big traders are probably not at the exact same price.
2/ Big traders have near 0 spread but medium ones do have spead, so the same stop entry at the very same place would not enter the market at the very same price for both of them.
3/ More important, you write yourself "The recent range is very small as limit orders are placed just before an SR line." Therefore not on the line. This is enough to turn...

After spending years watching price action the price will hit the same exact point on a chart to the very pip. Now I'm not saying this is 100% as nothing with so many variables can be but it happens enough to be uncanny. The reason that most traders use a buy or sell stop just prior to the SR line is the fact they want to insure it gets hit. I use five pips before but I don't buy or sell at the hit. I let the PA go through and reverse back. That way if the PA is blasting through the SR line to the next SR line I'm still in the trade.

As far as trendline use, I suggest using an indicator to verify after a PA cross. The use of trendlines with the SR's are a very powerful tool.

Johnny
  • #14
  • Feb 7, 2013 5:40am Feb 7, 2013 5:40am
very usefull
Mean Reversion
  • #15
  • Aug 16, 2013 8:49am Aug 16, 2013 8:49am
I kept searching. I realized that the number of times the price hits a level isn't the most important. Say we have a tweezers top candlestick pattern for instance, the price hit the high twice within two bars. We almost consider it a single touch. Now if we have a double top the price also hit the same level twice but it took much more time to retest the level. We will pay more attention to the double top even if the price levels aren't exactly the same. I noticed that my eye isn't attracted by where the price was but where the price WAS NOT. That is where the price was unable to go. We also notice the flat zones where the price knocked on a wall and failed to break it or it did only after a lot of attempts. Here the number of touches are important because of the time spent to generate them.
Attached Image

Our eye immediately sees the big picture that's why we are more interested in the blank space surrounding the peaks rather than the peaks themselves. Like Simon said in the first post fractals will happen at not-so-important areas. If we see a strong down trend we discard the inner pullbacks inside it. We'll get interested in them only to adjust the line, not to find it.
Attached Image (click to enlarge)
Click to Enlarge Name: exampleSwing.png Size: 94 KB

So I searched a way to "blur" the chart to only get the main path of the price. Blurring a function to remove the smallest peaks is a routine task in image processing. I used some sort of a morphologic closure. I project the highs and lows of a candle to the candles on the left and right if it can be reached is 1 or 2 bars distance.
Attached Image
No greed. No fear. Just maths.
  • #16
  • Aug 16, 2013 8:50am Aug 16, 2013 8:50am
I then search fractals in the result and attribute a score to them depending on how many bars cannot be reached in the past. I limit the score to 10 to not have too heavy weights. I don't consider the number of bars on the right side because the score of the most recent swings would vary over time and I want to get the lines in real time. Anyway the limit of 10 is quickly reached by the biggest swings and this doesn't seems to be an issue.

E/U weekly after blurring:
Attached Image (click to enlarge)
Click to Enlarge Name: EU_W1_Blur.png Size: 15 KB
No greed. No fear. Just maths.
  • #17
  • Aug 16, 2013 8:52am Aug 16, 2013 8:52am
Now I have a set of interest points I need to merge them to find the relevant levels. Here I inspired myself by another image processing method: the segmentation by regions merging. I start by considering each point as a cluster on its own. I iteratively merge the two nearest clusters until the distance gets bigger than a threshold. The resulting cluster is centered based on the relative weights. The heavier moves less while the lighter is "attracted" to the other one. To find the threshold I use the median value of the heights of the candles. For E/U weekly I find 130 pips.

The points are scattered in the clusters. I compute the standard deviation of each cluster. The levels are less significant when too mush "diluted". To graph the result I use a limit of 1/4 of the threshold. If the deviation is bigger the level is drawn as a zone with two dash lines above and below. Here is the current result:
Attached Image (click to enlarge)
Click to Enlarge Name: reult_levels.png Size: 25 KB
No greed. No fear. Just maths.
  • #18
  • Last Post: Aug 19, 2013 6:12am Aug 19, 2013 6:12am
Now I have a big problem to solve: how far shall I search the peaks? Usually 150 bars are enough. That's what I used to take the screenshots. But sometimes I need going way back in time. E.g. The recent high on EUR/AUD at 1.5000 is only matched by a resistance back in 2010. There is nothing before I don't want to systematically look back that far. It would include totally irrelevant levels. I'd like to find a way to detect the swings/trends/ranges I'm interested in including in the search. Not N bars backward but based on PA.

How does a human being scale his chart to draw the lines manually? What jumps to his eyes? When is he scrolling more in the past? Can I algorithmically mimic this?

Ideas are welcome.
No greed. No fear. Just maths.
0 traders viewing now