Pattern for signal

How do you know when a trend has started? Ended? This forum is for discussions about trend indicators and signals.
Post Reply
ratio
Roundtable Knight
Roundtable Knight
Posts: 338
Joined: Sun Jan 15, 2006 11:07 pm
Location: Montreal, Canada

Pattern for signal

Post by ratio » Mon Jan 29, 2007 10:49 am

Anybody had experience with using past market price pattern to generate signal.

I wrote this Program that basically calculate all the price pattern for a given length of day. ex: 2 day, and all relation of OHLC to the other days. HEre is an example of Corn on a 2 day pattern. The pattern is like that, It is the relation of
Open to Open
Open to High
Open to Low
Open to Close
High to Open
High to High
High to Low
High to Close
Low to Open
Low to High
Low to Low
Low to Close
Close to Open
Close to High
Close to Low
Close to Close

You can see the total of occurence, and the number of time the market was UP or Down after 1 and 5 day.

I found that some result are fairly high. Anybody had experience with that kind of a trigger.

I wrote that yesterday night, and I lost the source code, I have to rewrite it. What I want to do is have it produce the TB Entry Signal source code automatically.
Attachments
Corn.png
Corn.png (55.76 KiB) Viewed 5867 times

Forum Mgmnt
Roundtable Knight
Roundtable Knight
Posts: 1842
Joined: Tue Apr 15, 2003 11:02 am
Contact:

Post by Forum Mgmnt » Mon Jan 29, 2007 11:25 am

The problem with patterns as shown here is that you are only getting sample sizes in the 4 to 21 range for the individual patterns. This means most of these high percentage patterns have no statistical meaning. Your margin of error is too high.

Further when you have a computer do searching like this it will always find something in a small sample size, even of random data.

The sports guys are constantly making this mistake. They say well when the sky is blue, johhny has not eaten his wheaties, and they are facing Denver in october, then 100% of the time the Archies win. But they don't tell you that 100% of a sample size of 3 is meaningless. Especially when they have a computer search for all such occurences of the phenomenon.

The complex rules like the above are an example of overfitting to the data.

You need to test for multi-market patterns so you can get sample sizes in the 100s to start drawing any sort of conclusions with reasonable certainty.

- Forum Mgmnt

sluggo
Roundtable Knight
Roundtable Knight
Posts: 2986
Joined: Fri Jun 11, 2004 2:50 pm

Post by sluggo » Mon Jan 29, 2007 11:29 am

This sort of stuff has been around a long time. Perry Kaufman presented similar ideas in his book "New Trading Systems and Methods", twenty years ago. It might be worth investigating his approaches and results. The later editions include Tradestation source code.

However - beware of small sample sizes - keep in mind the astounding profitability of the Super Bowl indicator.
Attachments
pk2.jpg
From the 4th edition (2005)
pk2.jpg (86.88 KiB) Viewed 5857 times
pk1.jpg
From the 2nd edition (1987)
pk1.jpg (131.04 KiB) Viewed 5857 times

Post Reply