Machine Learning for Hedge Funds
Is there much interest in the recent developments in neural networks (deep, recurrent or reinforcement) at funds? Or are other method generally favored for financial time series prediction?
Trending Content
Career Resources
I've worked at several quant funds, including one that everyone would agree is in the top 5. I'm sure somebody somewhere is doing something with machine learning, but it just doesn't seem to be very widespread among the serious money makers. Neural networks will always find a solution to whatever training data set you give them and you can't ever describe it in words to see if there's any kind of pattern vs just random noise. I've never met a "just 100% trust the machine" trader in my life -- even the ones who most rely on algos still demand a human-understandable description of what the algo is looking for rather than just blindly trusting a black box.
Ideally, a neural net would help you find a historical correlation that you didn't know existed -- but the problem with neural nets is they always find a solution to any arbitrary set of training data you give it, and they don't give you any intuitive insight into why the pattern might exist, so it doesn't seem very fruitful.
Yes, but you can test the model's performance just like any human created algorithm. Just because a machine created it doesn't make it any more or less valid--although your typical trader might view it with more suspicion.
In fact, there are always going to be people who don't trust machines, just like there are people who still insist records or tube amps have better sound. Heck, I even heard there are people in Pennsylvania who still use a horse and buggy to get to work.
The real question is do modern machine learning algorithms, when supplied with the appropriate data, yield predictions that produce a positive return? If so, which ones? If none do yet, then I see a business opportunity (but I doubt that's the case).
I think morgan's point is that machine learning algorithms are more likely to produce spurious correlations as opposed to identifying patterns that will reoccur out of sample. However, I am not a quant, so ymmv
From my experience, more value comes from providing the modeling procedure with interesting inputs than in using increasingly complicated modeling procedures. That's not to say that there isn't value in using something like neural nets, but the incremental benefit is often marginal relative to the additional complexity they add.
The other thing is that with financial data the signal to noise ratio is typically a lot lower than in more traditional applications of modern machine learning, such as in image recognition. So the problem is inherently more difficult and you can't just easily throw tons of data at it since financial data is non-stationary at any time-scale you might look at it.
Recently there was an interesting WSJ article on Two Sigma's approach to quantitative investing. I do not work there, so I do not know precisely what they're doing, but I imagine that even though they're using computers to crunch tons of data, the actual statistical methodology behind it would not be very complex.
I've seen job postings for exactly these types of positions before, so I'm sure there's some interest. I have also heard from many that this is part of the system RenTec uses. Take that for what it's worth.
Do you study Machine Learning? Would be interested to know how widely is it used/ how valuable is it on the street since I can pick it as my fourth year course option.
Machine Learning Based Hedge Funds (Originally Posted: 01/23/2009)
I am graduating with an MS in Computer Science with a specialty in machine learning.
I have been looking for firms that specialize in machine learning strategies.
I have found RGM Advisors and Boronia, anyone know of other firms that specialize in machine learning?
Hank Greenberg's grandson, Spencer Greenberg runs a machine learning hedge fund called Rebellion Research :
http://online.wsj.com/article/SB100014240527487038346045753653108139480…
Dolores nesciunt dolores eum maxime et repellat. Rem consequatur consequuntur reprehenderit. Aliquid facere suscipit qui voluptate voluptas. Nulla voluptates sed consequatur illo.
Odio architecto porro placeat earum ducimus. Praesentium quia dolor molestias dolorum ipsam libero sit sint. Aut quod perspiciatis voluptatem. Quam nulla distinctio itaque eius.
Rerum eos modi sequi. Rerum in ratione voluptatibus dolorum fuga quibusdam expedita. Rerum maiores dolor sit sint cumque aspernatur enim accusamus.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...