Look Who it Was Random video

Hurdles for Machine Learning in Finance (w/ Dr. Dario Villani) | Discoveries | Real Vision™

2,988 views · Published 15 September 2018 · 3:12 · Indexed 8 November 2026

Channel: Real Vision Presents · 2018 · News & Politics

Watch on YouTube

Dr. Dario Villani, CEO of Duality Group and the former global head of portfolio strategy and risk at Tudor Investment, explains the major hurdles and opportunities companies face when implementing machine learning. While machine learning is the backbone of many of the world’s biggest projects, from self-driving cars to the development of virtual assistants, Dario explains why it has yet to be successfully applied to the world of finance and investing.

The only financial TV that matters. Start a 14-day free trial on Real Vision to watch more interviews like this: https://rvtv.io/2CK4Xd1

Watch more Real Vision™ videos: http://po.st/RealVisionVideos
Subscribe to Real Vision™ on YouTube: http://po.st/RealVisionSubscribe

About Discoveries:
Stretch your mind with brilliant original thinkers. This series introduces you to financial experts who bring a different angle to an existing topic or discuss ideas that are still on the boundary of finance. Challenge your existing thinking and be exposed to breakthrough topics that remain outside the realm of mainstream media. 

About Real Vision™:
Real Vision™ is the destination for the world’s most successful investors to share their thoughts about what’s happening in today's markets. Think: TED Talks for Finance.  On Real Vision™ you get exclusive access to watch the most successful investors, hedge fund managers and traders who share their frank and in-depth investment insights with no agenda, hype or bias. Make smart investment decisions and grow your portfolio with original content brought to you by the biggest names in finance, who get to say what they really think on Real Vision™.

Connect with Real Vision™ Online:
Linkedin: https://rvtv.io/2xbskqx 
Twitter: https://rvtv.io/2p5PrhJ

Biggest Hurdles for Machine Learning in Finance (w/ Dr. Dario Villani) | Discoveries | Real Vision™
https://www.youtube.com/c/RealVisionTelevision

Transcript: 
Machine learning hasn't really delivered in terms of incredible success in finance as much as it did in image recognition, video compression, speech recognition. Because there were, like, stumbling blocks and so the vision is very much helping that direction to remove those stumbling blocks, and having a chance for finance people to enjoy as much success as people on the West coast have enjoyed using machine learning. The problem of finance is particularly hard. You are dealing with objects that are statistically embedded in a lot of noise, that change all the time, because markets are not stationary in presence of relative scarcity of data. In terms of complexity, dealing with markets, it's much harder than recognizing a cat from a muffin, having 500 million images. So a lot of the interesting work that's being done, it will be done inside of financial institutions. Some of these issues are really profound and deep. I give you some example. Markets are reflexive. The simple act of learning something about the market dynamics and deploying capital to it, changes the dynamics of the market itself, creating positive and negative feedback loop. Such that, over time, noise can become signal, and signal can become noise. 
That's not the case for observing or recognizing cats, or distinguishing cats from muffins. Markets have very low signal to noise. No matter what you do, the noise is very dominant. And the best you can hope for is to capture tenuous, elusive relationships and make money out of it. And the other theoretical stumbling block was how do you deal with the relevance. And that's something that it's not present in image and speech recognition. It was really the challenging point that has prevented a lot of people from enjoying that success I was talking about. Well, the issue is an aggregate. It's a much safer endeavor than the one of discretionary trading or systematic trading. Of course, like, during Brexit, there are people who may end up having the right model or the right trade for the wrong reason, the right reason. And they're going to do very well. In machine learning, architecture will be very tentative. The yellow bus would go through the roof. And the machine would say, I don't have enough evidence to make a decision. I'm not going to just take a punt and try to get lucky
So in individual instances, like, it could outperform. But an aggregate is a much better approach, to be much more contemplative in nature and assessing all the probabilities and being tentative at the regime shifts or whenever there is a data break.

More from this channel