Thank you!
Machine Learning Trading Algorithms
- Thread starter userque
- Start date
I really like the basic idea of a pattern matching algorithm coupled with back test of results. Not sure yet which question I'd then need answered though. In addition to Larry Williams using something similar, I'm also reminded of Bulkowski ( http://thepatternsite.com/ ). Although he tries to backtest specific theoretical patterns of his own creation and not from current PA, ideas look similar, with a bit different approach.
Seems you don't necessarily need a kNN or NN at all, depending which route one would want to go to investigate / explore such ML trading algos.
I also totally agree that training an NN to find such patterns, seems like a wasteful approach and possibly too indirect. NN people will probably disagree, which is fine. This approach may sound simple, but there's still infinite approaches on how to solve even this.
Which is good, because if everyone traded the same, the markets just wouldn't function properly.
Seems you don't necessarily need a kNN or NN at all, depending which route one would want to go to investigate / explore such ML trading algos.
I also totally agree that training an NN to find such patterns, seems like a wasteful approach and possibly too indirect. NN people will probably disagree, which is fine. This approach may sound simple, but there's still infinite approaches on how to solve even this.
Which is good, because if everyone traded the same, the markets just wouldn't function properly.
I looked him up after you last mentioned him...I didn't see any information on him using pattern matching...I'll check more closely.
The main dissimilarity between mine, the immediately above, and (I suspect) the above, is that my system hunts out hidden patterns--rather than just obvious/visible ones.
I'm not sure. I created it. Then I later discovered that what I created was actually called a kNN algo. After some more research, I was unable to find any other traditional algo that I thought would do the job better. But, I also use a custom algo I created to deal with the curse of dimensionality.
Regarding Larry, hes mentioned he has people creating pattern matching algos for him, but I suspect hes not selling those and its entirely private.
Yes, Im thinking about visual patterns, not permutations of hidden ones, which is probably why your algo is so cpu intensive.
Yes, being a systems dev myself Its funny how you can make something out of iterative necessities, then later discover there are theories and papers for that stuff. Often the papers are so elaborate as to almost be useless for understanding and practical purposes.
Right. That, and the internal backtesting...and the ensemble of multiple kNN's.
Btw if you are interested in different types of seasonal patterns, Larry has made quite many webinars / lessons on that. Its one of the main stuff he teaches.
Haha. That's not the chart I was referring to.I'll repost an updated chart later today. You might have missed it.
Actually, I forgot the chart I posted listed the number of trades on it. That's why mistakenly thought you were looking at the 'wrong' chart. My bad.
Thanks for your answer of '30 trades.'
I'm fascinated by the way some brains work.
I was/am curious as to how long must a system, any system, work well before an outside observer--with no knowledge of the internal workings--would simply say, in effect, if not in these exact words, "Yep. It works!"
I understand some system may experience a run of 'luck,' hence my original question.
I guess customers on C2 must face this dilemma:
1. Do I hop on board a newer system that's doing well right now? And just hop off if the wheels start to come off.
2. Or do I wait for x number of days (x being the subject of my original question) for the system to 'prove' itself, all while missing out on great gains during the wait?
For me, performance is not a deciding factor regarding a system. How and why it works should be understandable, reproducible and logical. If truly met, I don't even need a backtest or proof. Visual graphing do help though, both in deciding what to throw away, categorize subsystems and what to develop further.
My intention from the beginning has been to "understand the market". Along the way, trying to become profitable has become a gold standard toward that goal. Otherwise, I could be just fooling myself, which is uninteresting and superficial. Especially if I manage to convince myself and others, but without some real benchmark to support my progress.
I suspect the mindset is entirely different for someone who is just after the money / income. This is true for most everything in life. The intentions and mindsets differ, although people can do almost identical actions. Funny thing is this can subtly affect the outcomes with dramatic differences, and sometimes not.
Thanks.
Sounds like you're saying that you would never invest your money in a fund, regardless of the (lack of) risk and (huge) returns; unless you totally, or almost totally understood how it worked (or its premise)? (I'm talking a hypothetical fund).
Your intent is to 'understand the markets.' I think I had that same intention many years ago.
There are many things we don't fully or nearly fully understand in this universe, yet we do capitalize off of them. Gravity is one.
Me personally, I'd rather spend time understanding how to make use of something, then, if still interested, understand how it works. Simply because there are only 24 hours in a day.
Also, I remember reading a while back the the markets are deemed a chaotic system. A system such that its inner workings are either unknown, or are unknowable.
I think I read this on a wiki page somewhere.
I accept that definition.
I am following this thread.