These trades generate around 80% of the total operational costs involved in trade processing. Delays in the trade matching process can also lead to regulatory breaches and trade settlement failures, as well as errors in valuation.
Among the many potential applications of AI and machine learning, the biggest potential benefits may lie in the automation of labour-intensive middle-office processes, and in particular the proactive prevention of blocked trades. BNP Paribas Securities Services is about to put this theory into practice with the launch of “Smart Chaser”, a decision-support tool that takes some of the pain out of trade processing.
From identifying to proposing actions needed
Smart Chaser applies predictive analytics to identify the minority of trades that may prove problematic and need intervention. It predicts the time these trades will take to match and suggests smart email “chasers” to counterparties needed to address the structural or system-related issues that typically cause delays, speeding up their resolution.
The Smart Chaser algorithm can identify the subset of trades that are most in danger of failing altogether, suggest the reasons why, and propose the actions needed to rescue them, thereby ensuring middle office teams can allocate their attention where it is most needed.
100 different factors
The system is dynamic in that it continually refines the model it uses to make predictions, using historical data to learn and improve over time. The key to this is the ability to factor in numerous reasons why trades might fail, and then perform analysis to spot patterns that indicate a high likelihood of failure.
Historically it’s been relatively simple to track the success rate for each broker, for example, and use that as a rule of thumb for likely trade success. The Smart Chaser algorithm is hugely more subtle (and accurate), since it includes around 100 different factors, including the broker’s history, the time the trade was executed, the total value of the trade, whether it was part of a block or a single allocation, the geographic locations of the counterparties, and so on.
Over time it builds up a complex and nuanced picture of the likely trade matching success rate, with 98% accuracy to date. The predictive model is based on a “Random Forest” algorithm, which incorporates multiple different decision-making trees to generate predictive outputs. The model is updated daily using data from the past three months to adjust the weights allocated to different factors and hence continually improve the accuracy of its predictions.
An interesting aside is that early models used one year’s historical data, on the assumption that seasonality had a big impact on trade matching success. As it turned out, a rolling three-month dataset provided greater accuracy, in part because brokers refine and update their own systems and personnel much more frequently than once a year, actions that have much a bigger impact on the likelihood of trades matching than whether they were executed late afternoon on December 24th, for instance.
Better identifying risk probability
Clients are likely to see three-fold benefits. One is speed: it should accelerate the average matching time and reduce the overall number of failed trades. Another is efficiency, with the middle office team spending less time on trades that are unlikely to be problematic, and the system automatically generating appropriate responses to chase those that are.
Thirdly, the system will provide much more transparency on trades, ultimately allowing clients to determine key risk identifiers based on the probability of trades failing and to monitor those that are most likely to be at risk.
Eventually, a more accurate picture of counterparties’ and stakeholders’ performance can be built up that takes a variety of factors into account. Whereas today, for example, brokers might be scored on crude numbers showing the overall proportion of trades matched on time, data collected by Smart Chaser should allow like-for-like comparisons across asset classes, trades of differing complexity and other factors.
Going beyond trade processing
Phase 1 of Smart Chaser will predict the likelihood of trade breaks and the time to match, and providing smart email templates for streamlined communication with counterparties. The system will be accessible through an intuitive trade-blotter-style dashboard that displays all trades to be processes and visualizes those at risk and in need of attention.
That’s just the beginning: Smart Chaser phase 2 will use the same methodology to scan emails received from counterparties notifying of problems in trades to identify the type of issue raised and generate an automated response. Phase 3 will be to combine the two, creating a complete trade-processing AI that will be able to predict and handle a wide array of problems with ever greater speed and accuracy.
Of course, the principles of predictive analytics and machine learning can be applied across a range of middle-office functions beyond trade processing. Reconciliations are one obvious area where advance identification of problems, analysis of their root causes and a degree of automation in their resolution would greatly increase efficiency and lower costs for our clients. That’s not to mention the possible applicability of similar systems on the sell side of the business. Ultimately we will all benefit.
Conscious of this potential we are deeply committed to developing Smart Chaser further, and urge clients to join us on this journey to enhance competitive advantage through the application of cutting-edge technology.