Gamestudio Links
Zorro Links
Newest Posts
Accessing All Historical Price Data
by JohnJits. 10/22/19 22:37
file_write var array ?
by laz. 10/22/19 16:00
SAV_STATS
by MatPed. 10/22/19 13:57
Instrument Price Zorro Chart
by jcl. 10/22/19 10:39
Calculating with price series
by AndrewAMD. 10/21/19 14:31
Error E355: Startup failure
by Dooley. 10/21/19 06:47
Today real live account blown - my experience
by Clifford_B. 10/19/19 17:09
AUM Magazine
Latest Screens
The Space Between
Pogostuck: Rage With Your Friends
Worst Case Z
AckCon'18 - Lotter vs the World 2 - Preview Release
Who's Online Now
4 registered members (JohnJits, AndrewAMD, Clifford_B, 1 invisible), 627 guests, and 2 spiders.
Key: Admin, Global Mod, Mod
Newest Members
VillageTalkies, line7, Shutch, Clifford_B, Paulius
18310 Registered Users
Previous Thread
Next Thread
Print Thread
Rate Thread
Page 1 of 2 1 2
Always accept detrended result as more realistic? #465844
05/15/17 05:42
05/15/17 05:42
Joined: Feb 2017
Posts: 358
D
Dalla Offline OP
Senior Member
Dalla  Offline OP
Senior Member
D

Joined: Feb 2017
Posts: 358
I was developing a trend following system the other day, and it got surprisingly good results. Eventually realized that I had forgotten to detrend the trades while training (but not while testing), and when doing that the system was completely destroyed (still profitable, but not comfortably so).

So should you always accept the detrended result as more realistic?

Also, when in your development process should you detrend the trades?
I tend to do it as a part of step 6 if you follow http://www.financial-hacker.com/build-better-strategies-part-3-the-development-process/

Re: Always accept detrended result as more realistic? [Re: Dalla] #466372
06/12/17 08:46
06/12/17 08:46
Joined: Jul 2000
Posts: 27,033
Frankfurt
jcl Offline

Chief Engineer
jcl  Offline

Chief Engineer

Joined: Jul 2000
Posts: 27,033
Frankfurt
It depends on the system. If it's perfectly symmetric in long/short, detrending is not necessary. If not, detrended results are more realistic. Use detrending also for training, otherwise the training would be affected from trend artifacts.

It is unusual that you got good results when detrending for testing and not for training. I would normally expect the opposite.

Re: Always accept detrended result as more realistic? [Re: jcl] #466378
06/12/17 09:37
06/12/17 09:37
Joined: Feb 2017
Posts: 358
D
Dalla Offline OP
Senior Member
Dalla  Offline OP
Senior Member
D

Joined: Feb 2017
Posts: 358
Does it make sense to to detrend the test results also?
Currently I'm using this in scripts, so only detrend during training

if (Train) {
Detrend = TRADES;
}

Re: Always accept detrended result as more realistic? [Re: Dalla] #466385
06/12/17 14:00
06/12/17 14:00
Joined: Jul 2000
Posts: 27,033
Frankfurt
jcl Offline

Chief Engineer
jcl  Offline

Chief Engineer

Joined: Jul 2000
Posts: 27,033
Frankfurt
That's what I'm using too. But when the strategy is asymmetric, f.i. long-only, it's better to also detrend in test mode.

Re: Always accept detrended result as more realistic? [Re: jcl] #471671
03/13/18 19:13
03/13/18 19:13
Joined: Dec 2017
Posts: 129
Halifax, NS
K
kujo Offline
Member
kujo  Offline
Member
K

Joined: Dec 2017
Posts: 129
Halifax, NS
I have a coupe of questions on detrending:

1. As I understand, it's always a good idea to use detrending (Detrend=TRADES) in the Train mode. This way parameters are not affected by a trend if there is any. But when it comes to the Test mode it's better to use detrending only with asymmetric strategies? No need to use it with symmetric strategies like Workshop4? Am I right?

2. Second question is about using detrending wth trend strategies. What's the idea of detrending in the strategy that is based on trend leveraging? Maybe it makes sense because underlying trend (that is removed by detrending) is at a much higher timeframe than the trends identified by the strategy. But again, by nature a Trend Following system is supposed to take advantage of this trends. Why to remove them?

Re: Always accept detrended result as more realistic? [Re: kujo] #471678
03/14/18 09:06
03/14/18 09:06
Joined: Jul 2000
Posts: 27,033
Frankfurt
jcl Offline

Chief Engineer
jcl  Offline

Chief Engineer

Joined: Jul 2000
Posts: 27,033
Frankfurt
Depending on the strategy, detrending is not always a good idea. When it exploits some long-term trending effect, you don't want to detrend the trades. When it uses short-term trend or mean reversion, and is asymmetric, then detrending is normally good.

Re: Always accept detrended result as more realistic? [Re: jcl] #471687
03/14/18 23:29
03/14/18 23:29
Joined: Dec 2017
Posts: 129
Halifax, NS
K
kujo Offline
Member
kujo  Offline
Member
K

Joined: Dec 2017
Posts: 129
Halifax, NS
In the 4th chapter of The Black Book the mean reversion strategy is symmetric, as I understand. Could you please explain why detrending is used in Train but not in Test?

Also, could you recommend a book or any source of info regarding detrending concept?

Re: Always accept detrended result as more realistic? [Re: kujo] #471690
03/15/18 08:07
03/15/18 08:07
Joined: Jul 2000
Posts: 27,033
Frankfurt
jcl Offline

Chief Engineer
jcl  Offline

Chief Engineer

Joined: Jul 2000
Posts: 27,033
Frankfurt
The strategies in the black book are mainly for illustrating concepts. Detrending is used in Train mode not because the system would absolutely need it, but more for showing how to activate detrending in a particular mode only.

Most books that I know are listed at the end of the manual. I remember the Aaronson book mentions the concept of detrending, and some of the other books probably too.

Re: Always accept detrended result as more realistic? [Re: jcl] #471691
03/15/18 11:13
03/15/18 11:13
Joined: Dec 2017
Posts: 129
Halifax, NS
K
kujo Offline
Member
kujo  Offline
Member
K

Joined: Dec 2017
Posts: 129
Halifax, NS
I see, thank you!
I'll ask more general questions then. From your experience when it makes sense to use detrending in Train but not in Test?

Re: Always accept detrended result as more realistic? [Re: kujo] #471693
03/15/18 13:50
03/15/18 13:50
Joined: Jul 2000
Posts: 27,033
Frankfurt
jcl Offline

Chief Engineer
jcl  Offline

Chief Engineer

Joined: Jul 2000
Posts: 27,033
Frankfurt
Almost always, since the test should be done with real historical data. Exception is when you need explicitely a detrended test result for some reason.

Re: Always accept detrended result as more realistic? [Re: jcl] #471694
03/15/18 14:17
03/15/18 14:17
Joined: Dec 2017
Posts: 129
Halifax, NS
K
kujo Offline
Member
kujo  Offline
Member
K

Joined: Dec 2017
Posts: 129
Halifax, NS
I see. So, to sum up, the general guidance looks like this:
- It's recommended to use detrending in Train almost always, unless the strategy exploits long-term trend.
- It's not recommended to use detrending during Test, unless the strategy is very asymmetrical, i.e. long-only.

Does it make sense?

Re: Always accept detrended result as more realistic? [Re: jcl] #471695
03/15/18 14:33
03/15/18 14:33
Joined: Feb 2017
Posts: 774
Chicago
AndrewAMD Online
User
AndrewAMD  Online
User

Joined: Feb 2017
Posts: 774
Chicago
I also have a question:

Is DETREND intended for observing DETREND training's effects on testing, or is it supposed to be used for code in production as well?

That is, would there ever be a reason to train data with DETREND and use that code in production?

Re: Always accept detrended result as more realistic? [Re: AndrewAMD] #471717
03/16/18 06:50
03/16/18 06:50
Joined: Jul 2000
Posts: 27,033
Frankfurt
jcl Offline

Chief Engineer
jcl  Offline

Chief Engineer

Joined: Jul 2000
Posts: 27,033
Frankfurt
Yes. You normally want to trade with parameters that work independent of trend.

Page 1 of 2 1 2

Gamestudio download | chip programmers | Zorro platform | shop | Data Protection Policy

oP group Germany GmbH | Birkenstr. 25-27 | 63549 Ronneburg / Germany | info (at) opgroup.de

Powered by UBB.threads™ PHP Forum Software 7.7.1