Forums  > Books & Papers  > Deep Learning for Event-Driven Stock Prediction  
Page 1 of 1
Display using:  


Total Posts: 2
Joined: Feb 2018
Posted: 2018-02-15 15:48
Does anybody have experience with the approach below? I am very skeptical regarding everything news-related and was wondering if somebody has already found some deficiencies with the approach.


Total Posts: 566
Joined: Sep 2006
Posted: 2018-03-27 20:47
Putting in this thread because it is also about DL: Rama Cont is a very smart guy but what do you think of this article ?

It claims that after feeding a DL network with lots of orderbook data that it can predict short-term stock movements. It even works well on stocks that weren't in the training.

Jim Simmons is finalizing a c. 1000 page tome on Renaissance trading methods and is looking for a publisher. He was turned down by Wiley. They said he's taxing the attention span of their readers -- LongTheta


Total Posts: 10
Joined: Jun 2018
Posted: 2018-06-04 20:56
Its indeed a giant game-changer because its his first paper in this direction and the fact that it claims success in over 60% of the trades without fundamentals, and regular techniques for stocks that are not in the training seems like a giant breakthrough.

Cause it means there are universal laws of price formation applicable across assets.


Total Posts: 1168
Joined: Jun 2007
Posted: 2018-06-07 14:24
Well. Here is where the magic happens:
"We train the network to forecast the next price move from a vector of state variables,"

They won't get more specific than that. I really really hold Cont in a high regard. I still consider his book on pricing with jumps one of the best books on pricing in general. But I doubt that this stuff works for real.

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


Total Posts: 459
Joined: Dec 2008
Posted: 2018-06-07 20:44
Hi Alfa - would you elaborate on the support for universal laws of price formation across assets?

Getting price formation right seems less a risk than getting price deterioration wrong.

The events that undermine the strategy are likely to be more rare in training/validation set than confirming circumstances, but arrive nonetheless with disproportionate severity.

That 60% probability is not going to degrade gracefully.

Previous Thread :: Next Thread 
Page 1 of 1