Trader behaviour analytics
A European investment bank was looking to develop a comprehensive monitoring solution to assess traders’ use of Cancel, Correct and Amend (CCA)
Sourcing the input data
The highly fragmented structure of our client’s trading system architecture required the ability to source trade information from a number of disparate systems.
High volume caused by the requirement to assess each version of all trades booked across trading business.
We used the enhanced data analytics technologies to load, transform and normalise trades from each of the trading systems.
Designing the Key Risk Indicator (KRI) tests
We developed a series of algorithms constructed using the Lavastorm Analytics Engine in order to highlight to the Anti-Financial Crime (“AFC”) unit any suspicious behaviour by individual / groups of traders.
These algorithms used the following statistical methods to assess the complete client trade population:
- peer group clustering calculation and identification of 95 and 99 percentiles
- random forest/decision tree analysis
- k-means clustering
The algorithms generated went through a rigorous back-testing regime which applied the each statistical method to a set of real transaction flow data and a set of control data to test the efficacy of the models developed.
In addition to the KRI tests constructed in Lavastorm, we used Qlikview to develop a CCA dashboard.
The dashboarding solution provided AFC management with a portal to track CCA KRI statuses.
A robust and repeatable CCA solution delivered fast
The CCA dashboarding solution we developed allowed our client to comply with both regulatory and internal audit requirements.
The process allowed active monitoring of trader behaviour and provided the AFC team with granular information to allow them to investigate specific KRI breaches.