Deep Stochastic Machine Learning

[AlexanderMcFarlane]

Has anyone seen this new field of machine learning used in finance?

If so what was the underlying use:

• Regression / Classification of Market data e.g. vol, price, stoch. vol, pairs etc.
  
• Regression / Classification of Social Media ('big data' etc.) to assist in trading
• both

All I can find is one paper (badly) using a convolutional neural network on a price series to no real avail.

Having spent some time coding these networks I am positive they could pin down some underlying non-linear relationships and the low latency once they're trained is obviously good... ie. I'm sure a quant somewhere is using them - question is who, how and where?

 

[vertigo]

ask Renaissance Technologies.... they know a 'little' about machine learning...social media to assist in trading? ..... are you serious?

i have a background in machine learning and heres what i learned: don't read anyones paper if it mentions shit like 'regression' or 'decomposition' or 'classification' etc etc... what are you trying to do? what are you trying to regress? what you trying to classify? why? why do you think the underlying data prescribes to that?

if you want to use machine learning techniques, try using a kernel method -> john shawcross taylor has a fantastic book on kernels and how to use them. kernels are useful for volatilities because... volatilities come in clusters, i.e. the ATM volatility curve is not (2d) uniform and the volatility surface is not (3d) uniform.. a kernel can be used to transform the curve/surface into being uniform and to track those clusters down.. turn nonlinear into linear...

 

[TehRaio]

is there a lot of use for machine learning in quant finance or economics?

 

[Gavin L]

Machine learning and statistics is used more in the quant hedge fund sphere?

 

[AlexanderMcFarlane]

ask Renaissance Technologies.... they know a 'little' about machine learning...social media to assist in trading? ..... are you serious?

i have a background in machine learning and heres what i learned: don't read anyones paper if it mentions shit like 'regression' or 'decomposition' or 'classification' etc etc... what are you trying to do? what are you trying to regress? what you trying to classify? why? why do you think the underlying data prescribes to that?

if you want to use machine learning techniques, try using a kernel method -> john shawcross taylor has a fantastic book on kernels and how to use them. kernels are useful for volatilities because... volatilities come in clusters, i.e. the ATM volatility curve is not (2d) uniform and the volatility surface is not (3d) uniform.. a kernel can be used to transform the curve/surface into being uniform and to track those clusters down.. turn nonlinear into linear...

Totally serious, see this link. I'd be very surprised if HFT funds weren't using NLP on Twitter and other social media to check for big volatility jumps.

From A Twitter Hack To The Complete Evaporation Of All Market Liquidity In One Chart | Zero Hedge

I don't for a second think that anyone in Sell side quant would really have much use for them though other than perhaps what you mentioned. I'm more asking about HFT and algo funds as the other guys mentioned