Velocimetrics has integrated machine learning into its market data quality solution, which was introduced about 18 months ago and is designed to accurately assess the timeliness and correctness of market data on a real-time basis. The addition of machine learning automates the task of categorising instruments in terms of what ‘normal’ should look like for each instrument, and supports the solution’s ability to detect and alert users to abnormal patterns among the instruments.
The enhanced functionality initially covers market data inputs to automated equity and foreign exchange trading, although Velocimetrics is planning to adapt its machine learning technology for a client working in futures and options, and says it will support other asset classes in response to client demand.
The software has been tested by a number of European investment banks and is now in full production. The inclusion of machine learning automates tasks that were previously manual and required significant set up time and ongoing administration, and makes the market data quality solution a more plug and play option providing real-time actionable insight at the single instrument and field level.
The software rebases what ‘normal’ looks like for thousands of instruments, each of which must be categorised depending, for example, on how regularly it should tick, what constitutes normal price movement, or how normal behaviour changes in response to specific market calendar events. Having learnt what normal is for all the instruments, the solution can detect and alert any anomalies and recognise the addition or removal of instruments.
Steve Colwill, CEO at Velocimetrics, says that in liquid markets, machine learning can build a sufficient history of an instrument within the market data quality solution in just a few hours. He explains: “Machine learning is often applied to Big Data after the fact. Velocimetrics’ innovation is in applying machine learning to thousands of live streams of data all the time. The mathematical concepts of the machine learning are similar, but the implementation is very different.”