The rise of machine learning in trading is posing new challenges for market surveillance, but the technology could also be a useful tool for identifying abuse risks, according to a report from the FICC Markets Standards Board (FMSB).
The increased use and sophistication of algorithmic trading and machine learning technologies is posing a significant challenge for the surveillance capabilities of firms, says the industry-led FMSB.
Firms now have huge amounts of structured and unstructured data but this creates the problem of noise, making it difficult to extract the data signals necessary to isolate and identify suspicious activity.
The increasing complexity of trading strategies and the nascent deployment of machine learning techniques also creates new challenges related to evidence of "intent, complexity, and the risk of self-learning machines actively choosing to manipulate markets," says the report.
But machine learning could also help market surveillance because it can process large complex data sets efficiently.
"Working side by side with humans, over time, machine learning programmes may be better able to understand the semantics of data and the evolution of behavioural patterns and to adapt their machine learning algorithms."
This means that market surveillance professionals will increasingly need a strong understanding of data science and technology in order to specify and test machine learning functionalities.
Read the full report:
Download the document now 1.2 mb (Chrome HTML Document)