Machine Learning at QB
Almost one year ago we launched our Blog with an article on Artificial Intelligence (AI) at QB. We wrote how AI is manifested in our execution algorithms, such as our flagship strategy, Bolt, which is a dynamic, intelligent algorithm that can make complex decisions in real-time. We are now entering the Machine Learning (ML) realm which is a whole new dimension and one we are rather excited about.
First, let’s take a step back and define what AI and ML are. They are often used interchangeably and are certainly not the same, but rather ML is a subset of AI. We like these easy to read definitions by Bernard Marr, an author and Forbes contributor:
“Artificial Intelligence (AI) is the broader concept of machines being able to carry out tasks in a way that we would consider “smart”.
Machine Learning (ML) is a current application of AI based around the idea that we should really just be able to give machines access to data and let them learn for themselves.”
Our Research team has recently completed a project on ML for Smart Order Routing in US Cash Treasury markets. This new application of AI has been built for the next generation of our Smart Order Router (SOR), facilitating intelligent aggregation of multiple liquidity sources. It is being used to solve the problem of where to send child orders when there is a choice between multiple liquidity pools. This is a something we have never faced before in futures, where vertically siloed exchange monopolies have prevented fungibility and therefore the viability of rival products, ensuring there is only one place to trade. Cash Treasuries is a different and fragmented landscape, consisting of numerous central limit order books and direct pricing streams. This results in considerably more complex market structure that we aim to solve for. Aggressive orders need to be routed to where the cost is the lowest (best price and minimum transaction fees). Passive orders are much more challenging – requiring us to solve a problem as to how to divide between venues in such a way as to maximize the expected passive fill quantity.
The good news is that our ML technology can figure out the solution to this problem in real-time. Our SOR, combined with our algorithmic strategies, create a powerful solution for the optimal execution of cash treasuries and improved efficiency for traders.
As Greenwich Associates asked in a recent blog article about treasuries, “where are the algos?”, the answer is right here at QB. This is the cutting edge of the evolution in Cash Treasuries market structure.
Quantitative Brokers
New York
June 27, 2018