Identifying a Price Action Pattern - Examples Included

[parliament718 10]

Hi,

Im trying to code a system based on a specific price action pattern. However Ive been trying a few days to accurately identify this pattern in code with little success. I can spot them out with the eye without a problem but I cant seem to figure out how to spot them in code because they come in slightly different shapes. Can somebody help out, some pseudo code or suggestions on how to approach is would really help.

 

Basically, the tunnel lines are just SMA's of the high and the low. Im looking for price to form this pattern and then close below the lower SMA for a short, or above the higher SMA for a long. The swings have to be well-defined (rather than "flat" waves, characteristic of consolidation zones). One of my attempts was to filter out these "flat" waves from the well-defined ones was by using an ADX filter.

 

Its most reliable when (for a short) during the 2nd wave the price comes out of the tunnel at the top (doesnt have to close above it) and then retreats back into the tunnel towards the bottom, and then closes below the bottom SMA.

 

There are of course other conditions and filters but this is somewhat the base of the system and Im having much trouble. Here's a compilation of examples.

 

 

Thanks so much

Edited May 14, 2012 by parliament718

[Tams 20]

Hi,

Im trying to code a system based on a specific price action pattern. However Ive been trying a few days to accurately identify this pattern in code with little success. I can spot them out with the eye without a problem but I cant seem to figure out how to spot them in code because they come in slightly different shapes. Can somebody help out, some pseudo code or suggestions on how to approach is would really help.

 

Basically, the tunnel lines are just SMA's of the high and the low. Im looking for price to form this pattern and then close below the lower SMA for a short, or above the higher SMA for a long. The swings have to be well-defined (rather than "flat" waves, characteristic of consolidation zones). One of my attempts was to filter out these "flat" waves from the well-defined ones was by using an ADX filter.

 

Its most reliable when (for a short) during the 2nd wave the price comes out of the tunnel at the top (doesnt have to close above it) and then retreats back into the tunnel towards the bottom, and then closes below the bottom SMA.

 

There are of course other conditions and filters but this is somewhat the base of the system and Im having much trouble. Here's a compilation of examples.

 

 

Thanks so much

 

 

We can make this a community project.

 

can you pick one pattern,

 

then, use point form, name 5 characteristics about this pattern?

 

they should be the 5 critical characteristics of this pattern

ie they are your trigger pullers.

use arrows and lines and poiinters and notes to supplement your descriptions

[parliament718 10]

Thanks for your reply.

 

I think part of my problem is its hard to quantify the pattern, maybe this is actually where I need help rather than the code itself. The schematic below of some solid characteristics. Please keep in mind this particular occurance is basically a perfect setup and they rarely happen like this.

 

 

1) L-H-HL for buy H-L-LH for sell

2) second wave typically retraces to 38%-78% fib before reversing again in the initial direction and closing over the SMA.

3)SMA crossing bar is "strong" but not so strong that it causes to miss much of the move. A range of the crossing candle's body of between 7-14 pips is a sufficient condition)

 

Since it comes in many shapes as seen in the previous compilation its hard to say other solid characteristics about the actual waves without making the definition so rigid that it misses perfectly valid setups. Perhaps theyre just less obvious. Here are some observations:

 

I just tried for the first time to run Fibonacci time retracements of the first wave. After looking back at 50-100 occurances that my eye spots out as having a good structure it seems that:

 

1) if the initial first wave is 7-15 (15M) bars of time, then the low/high of the following wave is usually in the 2nd - 3rd Fibonacci time zone

2) if the initial first wave is 16- 24 (15M) bars of time, then the low/high of the following wave is usually in the 1st - 2nd Fibonacci time zone.

 

3) if first wave is less than 7 (15M) bars of time, the pattern is usually too small

4) if the first wave is more than 25 (15M) bars of time, the pattern tends to be too big.

 

Perhaps a range for the initial first wave is also called for since without such a condition the code can be vulnerable to signal on "flat" consolidation waves. As I mentioned, I had some success filtering these out by ignoring signals with ADX < 20. Perhaps ATR might be useful in this regard too.

 

Thoughts?

[parliament718 10]

Here's 2 examples of what I meant in the previous post.

 

[parliament718 10]

The last 2 were perfect examples so here are 2 imperfect ones so Im not cherry picking.

 

Possible suggestions to correct imperfect setups, and other subtleties. I hope not to curve fit much...

 

 

[parliament718 10]

We can make this a community project.

 

 

 

Seems like there's no interest in a community project like this. If anyone is interested to work privately with me to build a system based on this feel free to message me.

 

Thanks

[equtrader 10]

Pattern identification is extremely hard and algorithms that do that efficiently are at the core of some commercial systems. In general, TA patterns are difficult to identify because of the imbedded randomness in the main pattern. Thomas Bulkowski has developed a program that does this and he offers it for free. You may want to contact him for general guidelines if he is available:

 

http://thepatternsite.com/patternz.html

 

Another guy who has developed an identification algorithm but for price patterns is Michael Harris. Although his program doesn't do what you are looking for his ideas have broader application:

 

The Most Advanced Tool for Analyzing Price Action and Discovering Trading Systems

 

I have seen some other programs but their identification ability is very poor. Check this website for an example of a screener where you can include TA patterns and indicators. Maybe you can contact them and get some direction:

 

Stock Screener - Overview

 

To summarize, what you are looking for is not very easy. A professional programmer of trading systems may charge up to $5,000 for the code. I paid much more for much simpler things 10 years ago.

[parliament718 10]

Pattern identification is extremely hard and algorithms that do that efficiently are at the core of some commercial systems. In general, TA patterns are difficult to identify because of the imbedded randomness in the main pattern. Thomas Bulkowski has developed a program that does this and he offers it for free. You may want to contact him for general guidelines if he is available:

 

http://thepatternsite.com/patternz.html

 

Another guy who has developed an identification algorithm but for price patterns is Michael Harris. Although his program doesn't do what you are looking for his ideas have broader application:

 

The Most Advanced Tool for Analyzing Price Action and Discovering Trading Systems

 

I have seen some other programs but their identification ability is very poor. Check this website for an example of a screener where you can include TA patterns and indicators. Maybe you can contact them and get some direction:

 

Stock Screener - Overview

 

To summarize, what you are looking for is not very easy. A professional programmer of trading systems may charge up to $5,000 for the code. I paid much more for much simpler things 10 years ago.

 

Checking back on this post now your comment helped me find two very useful articles for pattern recognition. One uses an algorithm called the PXtract algorithm:

 

PXTract Algo:

http://www.cse.ust.hk/~leichen/courses/comp630p/collection/reference-4-6.pdf

 

http://www.olsen.ch/fileadmin/Publications/Client_Papers//200405-Omrane-PredictiveSuccessProfitabilityChartPatterns.pdf

 

 

If somebody is interested to develop a pattern recognition system based on the pattern I described, then contact me. I now develop in a .NET-based Complex Event Processing (CEP) framework and software suite by Deltix. I prefer to work on systems in collaboration and besides, those articles are a bit much to digest because my math is a bit rusty these days. Hope to hear!

[equtrader 10]

Checking back on this post now your comment helped me find two very useful articles for pattern recognition. One uses an algorithm called the PXtract algorithm:

 

PXTract Algo:

http://www.cse.ust.hk/~leichen/courses/comp630p/collection/reference-4-6.pdf

 

http://www.olsen.ch/fileadmin/Publications/Client_Papers//200405-Omrane-PredictiveSuccessProfitabilityChartPatterns.pdf

 

 

If somebody is interested to develop a pattern recognition system based on the pattern I described, then contact me. I now develop in a .NET-based Complex Event Processing (CEP) framework and software suite by Deltix. I prefer to work on systems in collaboration and besides, those articles are a bit much to digest because my math is a bit rusty these days. Hope to hear!

 

The first article talks about the accuracy of the identification algorithm but not about the pattern performance. Also the authors state something peculiar that "We have demonstrated how to automate the process of chart pattern extraction and recognition, which has not been discussed in previous studies." but such software is available for at least 15 years! Then they refer to Thomas Bulkowski as Thomas! "According to Thomas [14], there are totally 47 different chart patterns, which can be extracted from the time series data". This is very strange. I doubt their algorithm will work because they do not sound familiar with this subject. Bulkowski has done much more work than them long time ago for chart patterns and Harris has done the same for price patterns:

 

http://thepatternsite.com/chartpatterns.html

The Most Advanced Tool for Analyzing Price Action and Discovering Trading Systems

 

those two are all you need.

[kenhigh 10]

Checking back on this post now your comment helped me find two very useful articles for pattern recognition. One uses an algorithm called the PXtract algorithm:

 

PXTract Algo:

http://www.cse.ust.hk/~leichen/courses/comp630p/collection/reference-4-6.pdf

 

http://www.olsen.ch/fileadmin/Publications/Client_Papers//200405-Omrane-PredictiveSuccessProfitabilityChartPatterns.pdf

 

 

If somebody is interested to develop a pattern recognition system based on the pattern I described, then contact me. I now develop in a .NET-based Complex Event Processing (CEP) framework and software suite by Deltix. I prefer to work on systems in collaboration and besides, those articles are a bit much to digest because my math is a bit rusty these days. Hope to hear!

 

Would like to share idea on fractal pattern recognition. A lot of good traders are trading it at twitter