2nd Annual
Machine Learning in Quant Finance
21-22 November 2019 | London Marriott Canary Wharf, UK
Interested for next year?
© Copyright 2019 marcus evans conferences
CONFERENCE OUTCOMES
On 21st-22nd November 2019 leaders from quant teams from across the financial market came together to discover how real world application of machine learning techniques, such as reinforcement learning and neural networks, could be applied in practice to solve specific financial problems. One thing that all attendees agreed upon at the beginning of the event was that if a machine learning project is not working, it is better to realise this as soon as possible in order to use this failure to learn lessons which can benefit future attempts. With this accepted, attendees cast a critical eye upon the market to learn from previous failures and, in doing so, seek out the wins.
The meeting kicked off by looking at how financial institutions can build the data necessary to enable sophisticated machine learning capabilities. With attendees estimating that quant teams are spending around 25% of their time on data cleansing, it was clear that this was a very important topic to them. Indeed the next ten years will redefine how firms use data so sourcing and enabling usability of intelligent data was a major focus. Within the scope of usability, attendees considered how different ways of clustering data can impact results from machine learning. In fact, one bank is moving toward clustering asset data by behaviour rather than the product names as such pulling together cross asset data. This same bank suggested that cleaning noisy data sets is where some of the most innovative work is currently being done.
This discussion around data then set the foundations to look at the challenges of applying sophisticated machine learning techniques such as explainability, validation, and governance of neural networks and reinforcement learning. Attendees agreed that a major point when applying machine learning is whether the problem at hand is resolvable, as if not then no matter how sophisticated the tools that you apply to it are, it simply will not work. The other challenge to consider is whether the right objective has been selected for the machine learning tool. At the meeting attendees widely discussed the Starcraft case study, where the AI agent only managed to win once programmed to beat the number one spot. This initiated discussion around whether the financial market can build a similar agent that is built with one important purpose in mind such as taking down the next flash crash? Being able to pick a resolvable problem and, the right objective for the agent requires a very specialist quant. This quant must understand how to engineer AI, be well versed in statistics and have deep knowledge of the business. This is so rare that the market is calling such a quant a unicorn.
Having looked at resolvable problems in the financial market, the focus was then on the machine learning toolkit and finding the right tool for the problem. The favoured school of thought here was to keep this simple, opting for the simplest tool first to see if this works before moving onto a more complicated one. When it came to the more complicated tools, the most widely used were Libraries and Python however neither of these tools appeared to offer much innovation. Ultimately, the innovation was in how the tool was being used rather than what tool was being used. This led to more cutting edge discussion, offering practical studies on the use of quantum computing and AI in the energy trading market.
