FIX Trading Community events are always a useful barometer of FinTech developments and last week’s Nordic conference in Stockholm was no exception. MiFID II is still a hot topic but the consensus is that its implementation went more smoothly than expected and attention is now turning to new, transformational technologies, like Blockchain and AI.
What interests Corvil about Blockchain is the part we can play in ensuring distributed network performance is aligned with expectations. We now know that some of the infrastructure that underpinned the first wave of implementations, such as Bitcoin, where it can take many minutes to confirm a transaction, was not as good as it could have been.
More pertinent, however, was the discussion around AI. Many market participants are already applying AI to large data sets and using predictive analytics to inform more profitable trading strategies. What they are less clear about is the part it can play in other parts of their ecosystems, specifically around assuring the performance and security of trading infrastructure, a Corvil area of expertise
I was on a FinTech panel where I introduced delegates to three use cases:
A prime use case is helping firms to stay inside regulated trading limits, which are often mandated on a daily basis. The order-to-trade ratio (OTR) limits that have been rolled out recently across European trading venues are a good example. By analyzing data from previous trading days, machine learning tools can predict when a trading system is on course to breach today’s limit, so that alerts can be triggered to prompt corrective action in good time.
This is a topical discussion for the FIX community, which is working on ways to standardize market metadata and make more it available for AI applications. Different exchanges currently have completely different ways of disseminating this data today. Not all of them are easily machine readable.
If the goal is to use AI to keep the self-driving trading car safely in the middle of the road and away from mandated regulatory limits at the edge, then the car needs to know where the limits are. Standardizing market metadata could provide a machine-friendly way to describe them.
For large firms trading different instruments in different venues for multiple clients, there’s been an explosion in data that has tested the ability of human beings to stay on top of it. Automation and machine learning can help by providing intelligent assistance to do two things: firstly, you can apply machine learning to historical data to identify anomalies versus past results. Second, you can use it to troubleshoot and investigate when something goes wrong. Detecting the root cause of a business-impacting event and resolving potential issues more quickly can minimize the impact on the trading environment and business result.
This kind of enhanced AI Ops is effective in any IT environment, which is why Corvil has a growing user base than spans networks, operations and security. But with the stakes so high in electronic trading, it’s a field where incremental improvements can be particularly valuable.
Corvil collects a vast amount of data and metrics from the environments we instrument. Based on all of this data, intelligent algorithms can learn to discern the impact of multiple operational factors on trading outcomes. When there’s a specific anomaly, such as an abnormal drop in fill-rate, techniques from explainable machine learning can be applied to interrogate the algorithms and identify the root cause, whether it’s high latency, excessive message rates or some change in market trading conditions.
Corvil is already strong on using network analytics to monitor a user’s behavior within an IT environment, detecting suspicious activity that may suggest an identity has been stolen or compromised. Similar behavior-learning techniques have the potential to be applied to trading algorithms and machines to detect abnormal changes in behavior, that could reveal a malicious (or errant) software change or a system problem leading to erroneous trading. Behavior-learning is a sophisticated approach to anomaly detection that examines multiple aspects of an entity’s activity to identify distinctive features, and is a significant focus area for our machine learning development team here at Corvil.
As these example use-cases show, there’s no shortage of opportunities to fruitfully apply AI and machine learning to improve the security and performance of trading infrastructure. I would point you to our latest product launch, Intelligence Hub, if you want here-and-now examples of how it is already making a difference to Corvil’s products and customers.