Trading System Lab

[Quantus]

Definitely, dear Transatlantic99, grail suite was ahead of it's time and offered for a fair price. Pity that many beta testers of USO (universal system optimizer) never got a final working product. Could be still state of the art, at least for retail customers.

Thank you for your comments, trader3cnd. What APS also suggested: 'You also can look for situations when two or more patterns give a signal at the same time with a sum of trades greater to 100. This is a valid approach, since all patterns that occur simultaneously are based on the same bar formations, while one may have more bars and another less.'

Dear Melanie, yes.. you really should understand the tools, knowing their pros and cons.. still, the kiss-approach has to be considered.

Quantus

 

[good one]

Quote from melanie911:

...complex tools easily drive to complex overfitting... imo

This is hardly a reason to not use complex tools, considering the markets are some of the most complex things in the world.
The real problem is that while a smoothe equity curve may be more significant than simply looking for anything with good absolute returns...the harder you look at the time series, the more you increase your chance of of finding a meaningless system that produces a smooth equity curve. Since there is no shortage of data in financial markets I just don't see how you can use a GA to find something from scratch. Your doing optimization on thin air. If you had time series data on 400k people flipping coins with 1 dollar bets on heads you can surely find a smoothe equity curve if you look hard enough. You could probly find an equity curve you want, what you really care about is the 399,999 that didn't produce anything.

http://en.wikipedia.org/wiki/Testing_hypotheses_suggested_by_the_data
http://en.wikipedia.org/wiki/Type_I_error#Type_I_error

 

[intradaybill]

Quote from jdeezero05:

This is hardly a reason to not use complex tools, considering the markets are some of the most complex things in the world.
The real problem is that while a smoothe equity curve may be more significant than simply looking for anything with good absolute returns...the harder you look at the time series, the more you increase your chance of of finding a meaningless system that produces a smooth equity curve. Since there is no shortage of data in financial markets I just don't see how you can use a GA to find something from scratch. Your doing optimization on thin air. If you had time series data on 400k people flipping coins with 1 dollar bets on heads you can surely find a smoothe equity curve if you look hard enough. You could probly find an equity curve you want, what you really care about is the 399,999 that didn't produce anything.

http://en.wikipedia.org/wiki/Testing_hypotheses_suggested_by_the_data
http://en.wikipedia.org/wiki/Type_I_error#Type_I_error

While I agree with what you are saying in principle, a coin flipping sequence that produces a smooth equity curve is based on independent tosses with two outcomes that have equal probability. In market data series you have more outcomes, like an open, a high, a low, a close, volume, open interest. The highs and the lows especially skew the statistics and introduce anomalies that can be detected by data-mining. A fundamental reason for this is that while coin tosses are independent experiments, market price moves are not because traders base decision on previous outcomes.

 

[good one]

Quote from intradaybill:
A fundamental reason for this is that while coin tosses are independent experiments, market price moves are not because traders base decision on previous outcomes. [/B]

True....I should add I namely trade futures off the tape + price levels and algo stuff is something I'm just working on but I doubt I will autotrade anything in the next decade...However obviously I don't discount past data to be "useless."
While of course what I posted doesn't map 1:1 there is an aspect of that with the markets...otherwise smart people wouldn't have bothered so much with markov stuff and markets.
What I philosophically don't understand is how you can start with nothing and then optimize "nothing"...Now if you started with something, then used optimization techniques to arrive at a better something I would get it...I just fail to see how starting with nothing then running massive GA leads to anything other than massive data snooping bias.

 

[intradaybill]

Quote from jdeezero05:

What I philosophically don't understand is how you can start with nothing and then optimize "nothing"...Now if you started with something, then used optimization techniques to arrive at a better something I would get it...I just fail to see how starting with nothing then running massive GA leads to anything other than massive data snooping bias.

Now, this is one of the most intelligent questions I have even seen in these threads. As it turns out, the predominant view at this point in physical science is that the universe started from nothing and that it is actually nothing since its total kinetic energy is equal to its total gravitational energy and when the two are added, nothing results.

You can get many profitable things out of nothing. Actually, the whole universe is a huge nothing.

However, I agree with you that snooping bias is a problem. I also dougt the effectiviness of GA and GP. I prefer data mining with simple search rules that restrict bias.

 

[Transatlantic99]

Quote from intradaybill:

I also dougt the effectiviness of GA and GP. I prefer data mining with simple search rules that restrict bias.

Counterpoint:

1) There have been several highly credible papers written on the effectiveness of GP in terms of overall robustiness of out-of-sample returns. This is in line with the "Adaptive Market Hypothesis."

2) I can state for a fact that several top quantitative hedge funds and investment bank prop desks make extensive use of GA's. And these are highly profitable desks. So you can argue that some of the top investment banks and hedge funds in the world are completely misguided and dont know what they are doing, but I dont think that stands up to serious scrutiny...

Transatlantic

 

[good one]

Quote from intradaybill:

Now, this is one of the most intelligent questions I have even seen in these threads. As it turns out, the predominant view at this point in physical science is that the universe started from nothing and that it is actually nothing since its total kinetic energy is equal to its total gravitational energy and when the two are added, nothing results.

You can get many profitable things out of nothing. Actually, the whole universe is a huge nothing.

Are you fucking kidding me? This may be the most bogus answer to a legit question I have ever run across on a message board..
good luck "intradaybill", that response puts you instantly on my "noise" list on this ridiculous board.

 

[good one]

Quote from Transatlantic99:

2) I can state for a fact that several top quantitative hedge funds and investment bank prop desks make extensive use of GA's. And these are highly profitable desks. So you can argue that some of the top investment banks and hedge funds in the world are completely misguided and dont know what they are doing, but I dont think that stands up to serious scrutiny...
[/B]

It should hardly be some suprise that the best in this game actually understand the concept of "Optimization"...gee imagine that..
Your so lazy though you don't even bother to understand how old hat GA is even JUST when it comes to the optimization of "something"..
I give up, this boards only utility is in studying Acrary post...nothing more.

 

[Transatlantic99]

Quote from jdeezero05:

Your so lazy though you don't even bother to understand how old hat GA is even JUST when it comes to the optimization of "something"..

Haha yeah I'm lazy. Good one. You obviously know me so well...

Love the way you can read into someone's entire background and knowledge of a subject based on one short post on ET. You are obviously very talented. Perhaps you should go into business as a mind reader or psychic. Good luck and be sure to keep us up to date with how that goes for you!

Transatlantic

 

[melanie911]

Quote from Quantus:


Dear Melanie, yes.. you really should understand the tools, knowing their pros and cons.. still, the kiss-approach has to be considered.

Quantus

yes

kiss-approach in trading too