Forums  > Basics  > turning points  
     
Page 1 of 1
Display using:  

nikol


Total Posts: 712
Joined: Jun 2005
 
Posted: 2019-06-11 09:08
What would be the easiest, yet efficient way to break OHLC historical time series into turning points?
Forward/backward search is allowed.
Of course, that will depend on the interval size and the allowed level of noise.

Thinking about the problem, I found couple of papers, here is one of them:
https://link.springer.com/article/10.1007/s00704-019-02817-9

~15 yrs back I did trend identification over tick data by using local forward looking linear regression with Chi2, but I realise that this problem can be reduced to the search of "turning points" only.

For example, linear trend going downwards is wrong for my purpose, because we would be interested in price going down (for short position), till price starts rising again (turning point).

Maggette


Total Posts: 1119
Joined: Jun 2007
 
Posted: 2019-06-11 09:14
CUSUM? No?

Ich kam hierher und sah dich und deine Leute lächeln, und sagte mir: Maggette, scheiss auf den small talk, lass lieber deine Fäuste sprechen...

nikol


Total Posts: 712
Joined: Jun 2005
 
Posted: 2019-06-11 09:20
Does CUSUM reset to zero when series of "greens" are interrupted with small "red" although the trend is not broken?

PS. will take a look, thanks

nikol


Total Posts: 712
Joined: Jun 2005
 
Posted: 2019-06-11 10:08
Understood

CUSUM(Index(some condition)) must work

Thanks a lot

Maggette


Total Posts: 1119
Joined: Jun 2007
 
Posted: 2019-06-11 10:49
Well, it depends. It workedvery well for me ouside of finance, where signal to noise ratio was quite high and the change in trend was quite obvious to detect.

Edit: hence I would be happy to hear / get some feedback from you how things turnded out for you.

Ich kam hierher und sah dich und deine Leute lächeln, und sagte mir: Maggette, scheiss auf den small talk, lass lieber deine Fäuste sprechen...

nikol


Total Posts: 712
Joined: Jun 2005
 
Posted: 2019-06-11 11:55
Sure.
I remember, you mentioned CUSUM ~6 months ago in a different thread. I use it now to detect buy/sell bursts (by counts and volume). It helps to identify moves across the network of liquidity pools.

nikol


Total Posts: 712
Joined: Jun 2005
 
Posted: 2019-06-12 11:59
Magette,

here is one of papers I looked at

"Cusum Techniques for Technical Trading in Financial Markets"

Talking about turning points, here is the result of changes of (close>open). Still need some cleaning. Plotted is 1-minute closes of BTCUSD somewhere in Jan-2019.



PS. I guess, that the final result will depend very much on timeframe of expected trade and minimum high-low range you allow.

Jurassic


Total Posts: 221
Joined: Mar 2018
 
Posted: 2019-06-12 12:09
Is turning points a big topic in time series analysis? I swear I never saw this topic in Times Series Analysis by Cryer (standard text on the subject).

Btw, is climate modelling a good place to look for new techniques which could be applied to financial markets?

nikol


Total Posts: 712
Joined: Jun 2005
 
Posted: 2019-06-12 13:05
@Jurassic

You can guess that there are many "topics" floating under carpet of traders research which are not taught by professors and not written in the books and will be not for some time.

PS. "I swear I never saw this topic in Times Series Analysis by Cryer (standard text on the subject)." - I consider it as a compliment. Thanks.

PSS. Climate modelling is not the only place to look at. I remember, back in ~2003-2005 biologists/gene analysts were very popular in finance due to their PCA. I saw nice methods in linguistics. Kantorovich back in 1939 advanced in linear programming while solving problems of optimal production in a plywood industry for which he was given Nobel prize in economics. Math is everywhere.

Maggette


Total Posts: 1119
Joined: Jun 2007
 
Posted: 2019-06-12 17:07
@nikol
THX

Ich kam hierher und sah dich und deine Leute lächeln, und sagte mir: Maggette, scheiss auf den small talk, lass lieber deine Fäuste sprechen...

Maggette


Total Posts: 1119
Joined: Jun 2007
 
Posted: 2019-06-12 17:07
@nikol
THX

Ich kam hierher und sah dich und deine Leute lächeln, und sagte mir: Maggette, scheiss auf den small talk, lass lieber deine Fäuste sprechen...

Maggette


Total Posts: 1119
Joined: Jun 2007
 
Posted: 2019-06-12 17:07
@nikol
THX

Ich kam hierher und sah dich und deine Leute lächeln, und sagte mir: Maggette, scheiss auf den small talk, lass lieber deine Fäuste sprechen...

Maggette


Total Posts: 1119
Joined: Jun 2007
 
Posted: 2019-06-12 17:08
@nikol
THX

Ich kam hierher und sah dich und deine Leute lächeln, und sagte mir: Maggette, scheiss auf den small talk, lass lieber deine Fäuste sprechen...

Maggette


Total Posts: 1119
Joined: Jun 2007
 
Posted: 2019-06-12 17:08
@nikol
THX

Ich kam hierher und sah dich und deine Leute lächeln, und sagte mir: Maggette, scheiss auf den small talk, lass lieber deine Fäuste sprechen...

Jurassic


Total Posts: 221
Joined: Mar 2018
 
Posted: 2019-06-12 17:37
@PSS. Climate modelling is not the only place to look at. I remember, back in ~2003-2005 biologists/gene analysts were very popular in finance due to their PCA. I saw nice methods in linguistics. Kantorovich back in 1939 advanced in linear programming while solving problems of optimal production in a plywood industry for which he was given Nobel prize in economics. Math is everywhere.

@nikol if math is everywhere then you need to find areas where there are interesting things going on!

nikol


Total Posts: 712
Joined: Jun 2005
 
Posted: 2019-06-12 18:11
E-commerce market making

ronin


Total Posts: 454
Joined: May 2006
 
Posted: 2019-06-14 08:57
CUSUM is basically a cliquet on the underlying process?

Surely there are more informative measures of non-Gaussianity. This could trigger because of autocorrelation, but it could also trigger because of kurtosis, vol of vol, jumps, and who knows what else.

"There is a SIX am?" -- Arthur

nikol


Total Posts: 712
Joined: Jun 2005
 
Posted: 2019-06-14 14:01
The subject is Turning Points, not CUSUM. I guess, that TP is model free (upto identification, filtering method). For me it is important to identify where they are. Even cheating with forward information is allowed.
If I have those TPs, then I plan to reverse engineer parameters of large pool of indicators in parametric space when t=TP (signal) versus t=/=TP (noise) and select the most sensitive ones.

ronin


Total Posts: 454
Joined: May 2006
 
Posted: 2019-06-14 17:41
More or less.

If your time series is Brownian, there are no turning points. They don't exist. You are just fooled by randomness, to borrow a phrase.

"There is a SIX am?" -- Arthur

nikol


Total Posts: 712
Joined: Jun 2005
 
Posted: 2019-06-14 21:51
I have never seen Brownian, but your argument of "fooled by randomness" is still valid.

Previous Thread :: Next Thread 
Page 1 of 1