Forums  > Careers  > Machine Learning/Deep Learning & HFT: Job Offer and/or Selling IP-Strategy  
     
Page 1 of 1
Display using:  

GeekySerge


Total Posts: 43
Joined: Apr 2010
 
Posted: 2017-11-24 16:13
Hi all,
after few years since my last thread in the career section and given the good amount of advice I've received in that occasion, which I greatly appreciated, I'd like to open a new one.

After starting my career as a vol trader for a well known US investment bank I've decided to take a career break to do a bit of research in applying forecasting models to HFT (machine learning/deep learning et al.) via an academic commitment. If you're not bored by the details there's a longer description in the addendum at the end of this post.

On paper I've obtained decent results (6+ Sharpe ratio after TC on 1min data and quite higher on 1sec/tick-by-tick, across asset-classes) so I'd like to move into production but I value the fact of working with other hopefully smart people and having a decent execution infrastructure, hence I've decided to look for a place where to work/implement my models.

A few days ago I've started making applications in a discreet manner, hence I've sent my CV to 1 company and talked with 1 HH. The HH seems interested in my profile but doesn't have an open position for me.

The company I have contacted (small prop shop mainly made of STEM PhDs with background in IBs/quant HFs) came back very quickly and we had few talks which came to be concentrated on my strategy. They seem very interested and they proposed different options. Among these, joining their team and/or buying my IP.

A further detail, in the past months I've been contacted by some HHs via LinkedIn for positions in a well known international HFT shop and a quant HF (started by ex D.E.Shaw people). Given that I didn't update my LinkedIn profile with my research I guess they have been hooked by the reputable names I have on it and by my academic background/skills (computer engineering/applied mathematics/financial engineering). Still I had to complete my academic commitment, hence I didn't pursue these opportunities (from the news I later realised that the HFT shop went into a big hiring campaign in the same period).

Given the above I'd like to have your opinion/advice, e.g.:
- Would you sell a kind of modelling framework like the one I've built? If yes, how do you value something like this?
- What do you think about joining a small team (to do ML research for HFT and implement my model) vs a bigger company?
- I like the people I talked to but I don't want to limit my options for not asking other places, hence should I apply to further HFT shops/quant HFs as well before committing to an offer?

Many thanks









PS:
I've read the (somehow) related threads:
http://www.nuclearphynance.com/Show%20Post.aspx?PostIDKey=167619
http://www.nuclearphynance.com/Show%20Post.aspx?PostIDKey=140649



ADDENDUM/DETAILS:
During my research effort I've built a decent automated modelling stack:
- data feed recorders
- data pre-processing/normalization
- feature engineering/variable selection
- (forecasting) models estimation
- trading filter from model forecasts to beat b/a costs
- model comparisons across asset-classes via Sharpe/PNL (after TC) calculated on out-of-sample virgin data.

To further reduce the possibility of introducing some bias (future data leak bias, selection bias/data snooping, etc.) somewhere in the stack I've implemented and improved different models from various academic fields (not only ML) using totally different coding libraries. In particular I had to spend a decent amount of time to improve the non-convex optimization procedure to estimate the parameters of some non-linear models (this is a somehow known problem for experts in the field which I solved for non-stationary time series).

All this has lead me to (probably) spot some interesting empirical facts which are recurrent across different models, with the more complex ones capable of beating the b/a costs. My preliminary tests show that such framework can be applied to lower frequency/higher capacity strategies too.
Previous Thread :: Next Thread 
Page 1 of 1