Forums  > Trading  > Deep learning for trading  
     
Page 1 of 1
Display using:  

Swissstud


Total Posts: 4
Joined: Aug 2008
 
Posted: 2016-12-16 15:26
There are workshops/summits, such as those from re-work coming up about deep learning / neural networks for finance, e.g. here:

https://s3.amazonaws.com/re-work-production/brochure_pdfs/47/original.pdf?1475763039

I was wondering whether anybody has comments on which tasks / problem setting and their corresponding DLL model / architecture might be appropriate.

I know it can be used for sentiment analysis, e.g. here: http://nlp.stanford.edu/sentiment/. I get the impression DLL do well with a well-defined problem structure and common application fields include speech recognition and computer vision.

Are there applications for position sizing, entries, exits as well?

mtsm


Total Posts: 179
Joined: Dec 2010
 
Posted: 2016-12-16 18:30
I can't predict what it will be like in the future, but the re-work workshop on the topic in the past was completely useless with regards to the questions you ask...


jslade


Total Posts: 1057
Joined: Feb 2007
 
Posted: 2016-12-17 01:30
I think it's a silly thing to attempt, but some kind of recurrent network (possibly with reinforcement training) would be interesting to look at. There isn't as much research published here as there is in feedforward architectures.
Position sizing and the rest: you have to model the probability distribution of your forecast versus reality. Neural approaches can be jury rigged to do this, but it makes more sense to model this some other way.

I agree with mtsm about the substance of the conference. Too much hype. Why aren't there conferences for using probabilistic graphical models in finance? Maybe because the people using them are making money rather than going to conferences.

"Learning, n. The kind of ignorance distinguishing the studious."

goldorak


Total Posts: 979
Joined: Nov 2004
 
Posted: 2016-12-17 14:17
> Maybe because the people using them are making money rather than going to conferences.

True for lots of topics. Blush

If you are not living on the edge you are taking up too much space.

finanzmaster


Total Posts: 94
Joined: Feb 2011
 
Posted: 2016-12-24 22:45
OK, if you want my comprehensive opinion, have a look at.
Big Data and Deep Learning, a technology revolution in trading or yet another hype?

But to put it briefly, I am pretty skeptical about the predictability of the market if you have only historical prices at hand. I do believe in a kind of mean reversion to fundamentals: ironically, I have recently bought a PUT on nVIDIA (whose GPUs are very popular for deep learning) because the stock growth is far exuberant compared to earnings growth. But if you want to match prices and fundamentals, you may (and probably should) use a more simple (and tractable) model.


www.yetanotherquant.com - Knowledge rather than Hope: A Book for Retail Investors and Mathematical Finance Students

katastrofa


Total Posts: 344
Joined: Jul 2008
 
Posted: 2016-12-25 02:20
Depends on what time scales are we talking about. There is quite a lot of predictability on the scale of few seconds. Different question is how much of it is tradable.

finanzmaster


Total Posts: 94
Joined: Feb 2011
 
Posted: 2016-12-25 10:29
@katastrofa, correct!
I meant daily and longer intervals

www.yetanotherquant.com - Knowledge rather than Hope: A Book for Retail Investors and Mathematical Finance Students

Rashomon


Total Posts: 165
Joined: Mar 2011
 
Posted: 2017-01-05 01:16
Kay Giesecke now has a paper using deep learning. I am not sure whether this should make me take DL more seriously or KG less.

"What you gonna dooo, without your ass?" ~ Sun Ra

agentq


Total Posts: 23
Joined: Jul 2008
 
Posted: 2017-01-06 06:33
This one?
link

For delinquency or prepayment modeling, this doesn't seem like a stretch.

mtsm


Total Posts: 179
Joined: Dec 2010
 
Posted: 2017-01-06 17:19
yes this one. it's on arxiv.

Rashomon


Total Posts: 165
Joined: Mar 2011
 
Posted: 2017-02-21 09:12
why use deep learning to model mortgage risk?

https://bfi.uchicago.edu/sites/default/files/file_uploads/Slides%20Giesecke.pdf slide 5
Previous Thread :: Next Thread 
Page 1 of 1