[MixedHerbs]

Hello,

I am about to start a hobby-project that will hopefully get a nagging idea out of my head. However before starting, I would like your views on whether it is completely daft. It involves:

1. Grabbing as much historical data as possible for a whole market.
2. Arranging the data in such a way that the whole database can be searched for a pattern formed by the last 15minutes (say) market activity.
3. Count how many times a scenario has taken place before - if at all.
4. Extract details of what happens next.
5. Decide if the odds of making money are good.
6. Search/calculation time to be less than a second.

Has anyone tried this before? I have searched the forum, but found little. I apologize in advance if I missed something obvious.

Regards, Peter.

[zdo]

Building an array of setups is a good idea but using pattern recognition has not borne the fruit most of its visionaries thought it would. Since the 80’s a series of commercial pattern recognition programs have come and gone. Just saw a blurb about one in a tv commercial (eTrade ?) recently. The practical application of pattern recognition has several limitations. One is the tolerance / fuzziness of the underlying templates - which in the trading world needs to be variable , looser then tighter the looser, etc.... But how does one intelligently pull that variability off. Second, patterns in a vacuum may show slight statistical odds, but when applied real time aren't so shiny. Third patterns not run in a vacuum, ie those with a contextual layer above the actual pattern require, again, variable tolerance in both the granular pattern and the ‘background’ pattern. Fourth, ….
What you’re thinking can be done. It’s a specialty, a niche… and if it's you, don't let anything discourage you... but bottom line, pattern recognition algos in general do not produce extraordinary results. hth

[Tams]

Originally Posted by MixedHerbs »

....Has anyone tried this before? I have searched the forum, but found little. I apologize in advance if I missed something obvious.

Regards, Peter.

Yes... many people have done this before...

One commercial service (the name escaped me at the moment) has their database and search software for rent at $5k per month.


when you say 15 min pattern, what resolution are you talking about?
at 5 min interval? 1 min? 1 sec? 1 tick?

have you worked out the factorial number of a 2 bar + volume pattern?

[MixedHerbs]

Quote from ZDO;

"Building an array of setups is a good idea..."

That's not quite what I meant. My intention is to arrange the data in such a way that it can be searched as efficiently as possible - probably using bitmaps. Before indexing, I would convert the raw data into a form that made volumes and prices the same scale. I would take the last period's data for a share price, convert it into the same form as the index, then work my way backwards to discover if history has repeated itself. As you say, there will need to be some adjustment for "granularity", but it will be fun to discover.

Searching for contextual layers is an interesting idea, thank you. I will put some thought into that.

Regards, Peter.