Spotting a trend

[mikesmithv]

Quote from logan04x:

Hi all, I'm working on a hobby software project to detect the trends and patterns inside a chart. This will keep me busy over the weekends and also educate me in technical analysis. And I had one question in this regard.

Currently I spot up or downtrend inside a chart by looking at the highs and lows. This works but it finds way to many lines, which causes noise. I want to know what are the ways by which I can filter out the noise from the real trend lines.

I have relied on the following heuristics for now:

1. Number of points on the line (should be >= 3)
2. Distance between the points

Can the experts give some more ideas on what indicators and factors should I look at to filter the noise from the rest of the lines.

I'm attaching a screenshot of my tool. As you can see that it finds many lines but I need to find the stronger ones out.

I too am a software engineer keen on UI who came into trading about 10 years ago and you remind me of myself back then. I wrote a lot of charting UI expecting to learn from it and after a while it became obvious that if my model for creating a trend line, for example, had any meaning beyond just looking pretty then it must have some kind of predictive power.

For example, if the line I plot off my data points indicate perfectly that there is a strong trend and the trend is up then what meaning does this have unless I can buy now and sell later for a profit? I did eventually learn a lot but it would have been a lot faster if I didn't feel like I had to out-do every chart programmer with a better, cleaner, more intuitive version of every chart I looked at and write my own version - like I could plot the points of a trend line in a way no one else thought of, my own special tweak that resulted in perfect predictions, ha!

In retrospect I would have been better off had I split my time between using off-the-shelf charting software like a "normal" person learning to trade and spent my programming time exclusively running simulations proving or disproving those trends and patterns using various strategies and getting a sense of the benefits and pitfalls of simulators.

Hmm, reading this I'm thinking I may be simply revealing what a lunatic I am. But if it has meaning I'm glad to help, lunatic to lunatic.

 

[logan04x]

Quote from mikesmithv:

I too am a software engineer keen on UI who came into trading about 10 years ago and you remind me of myself back then. I wrote a lot of charting UI expecting to learn from it and after a while it became obvious that if my model for creating a trend line, for example, had any meaning beyond just looking pretty then it must have some kind of predictive power.

For example, if the line I plot off my data points indicate perfectly that there is a strong trend and the trend is up then what meaning does this have unless I can buy now and sell later for a profit? I did eventually learn a lot but it would have been a lot faster if I didn't feel like I had to out-do every chart programmer with a better, cleaner, more intuitive version of every chart I looked at and write my own version - like I could plot the points of a trend line in a way no one else thought of, my own special tweak that resulted in perfect predictions, ha!

In retrospect I would have been better off had I split my time between using off-the-shelf charting software like a "normal" person learning to trade and spent my programming time exclusively running simulations proving or disproving those trends and patterns using various strategies and getting a sense of the benefits and pitfalls of simulators.

Hmm, reading this I'm thinking I may be simply revealing what a lunatic I am. But if it has meaning I'm glad to help, lunatic to lunatic.

Ha, thanks for the response . I'm actually using Python for this one (though I'm a C++ programmer at heart). The reason I choose Python is because it has all the goods of an easy to use language and a very powerful set of libraries. For drawing the charts I'm using matplotlib (here is an example of how easily you can plot candle sticks with it - http://matplotlib.sourceforge.net/examples/pylab_examples/finance_demo.html, you can also overlay volume or other indicators with a few lines of code). For me GUI is not so important as are the core classes which do the crunching of numbers to find patterns. However, GUI becomes important to see the results of the code and see if it found the pattern or not, which is very easy with matplotlib. Somebody also pointed me to ta-lib, which I will use for calculating a number of indicators as I move along.

So far I have coded up the basic pattern class heirarchy and added double top/bottom to it. Today worked upon head and shoulders top, which will be done by end of the week. Does all this work? Yeah, kind of, if the pattern is very visible while looking at the chart then the tool should also be able to spot it. I believe we can make it more better as I go along (like closely look at volume changes and accordingly give the pattern a rating for strength/clarity etc).

Thanks!

 

[logan04x]

Let me point out that I just want a tool that scans multiple charts and tells me if there are any emerging patterns or any patterns that completed the formation. I don't want to predict something that no one else knew about .

 

[ProfLogic]

Quote from logan04x:

Let me point out that I just want a tool that scans multiple charts and tells me if there are any emerging patterns or any patterns that completed the formation. I don't want to predict something that no one else knew about .

I have a tool like that and it was a bugger to create. It dumps the data to an Excel spreadsheet each morning telling me what charts are ready to immediately execute positions.

It doesn't predict, it reads. what is the difference you say . . .

A weatherman predicts it will rain tomorrow.

A "reader" stands in the middle of the yard, feels the strong winds coming from the direction os a major storm cloud he sees along with the smell of rain just a few miles away and sees the lightning and thunder associated with that storm and then makes the statement, "It's going to rain".

 

[deadbroke]

Quote from logan04x:

I'm new here and don't hang around a lot, but I wonder how someone with as few posts as you knows so much about Jack Hershey? I'm confused with some of the posts going around here .

Can anyone give me some background here?

Stay focussed on what you are looking for. Right click on put-down posts/send to/recycle bin.

Jack is a thorough gentleman whether you agree or disagree with his technical analysis.

Goodluck to you, Sir.

 

[logan04x]

Quote from ProfLogic:

I have a tool like that and it was a bugger to create. It dumps the data to an Excel spreadsheet each morning telling me what charts are ready to immediately execute positions.

It doesn't predict, it reads. what is the difference you say . . .

A weatherman predicts it will rain tomorrow.

A "reader" stands in the middle of the yard, feels the strong winds coming from the direction os a major storm cloud he sees along with the smell of rain just a few miles away and sees the lightning and thunder associated with that storm and then makes the statement, "It's going to rain".

Yes, I want it to read as well and not predict.

Quote from ProfLogic:

I have a tool like that and it was a bugger to create. It dumps the data to an Excel spreadsheet each morning telling me what charts are ready to immediately execute positions.

You mean it was too difficult to create it or maybe too easy? I'm curious to know more about it.

 

[ProfLogic]

Quote from logan04x:

Yes, I want it to read as well and not predict.

You mean it was too difficult to create it or maybe too easy? I'm curious to know more about it.

No, it was difficult to create because all oscillation outcomes had to be taken into consideration and on my charts there are 8 specific oscillation types. Of these 8 oscillation types at least 3 types, at any given moment, must activate while moving in or being in the same direction.