Overview
Since the financial crisis of 2008, banks have been under additional pressure to ensure that they are compliant with the regulations required for stress tests, with banks needing to ensure they have enough cash in reserve in the case of a catastrophic market failure.
In 2015, a large global Investment Bank approached DataSpartan looking to use reinforcement learning – the same type of learning that was used in game engines such as Alpha Go. They wanted an algorithm developed that could detect the “state” of the market and prompt traders to act accordingly based on historical data.
Approach
The bank had previously used another well-known consultancy to undertake the project to limited success. The consultancy had used templates that were designed in a one size fits all and nuances and intricacies in the client systems meant that a custom build was needed.
DataSpartan provided a custom build that was tailored to the nuances of the client data. Other inputs had to be considered, such as the Volker metrics in compliance with the Dodd Frank act. Custom APIs needed to be built to allow the in-house developers to build team specific functionality on top of the platform. Because cross-team data was required, a standard data capturing tool was built to ensure that teams provided the system with an input that was compliant and made data integration into the calculation engine much easier.
Results
The bank was highly satisfied with the work and is using the system internally for stress testing purposes. The system reports on tests on a daily and weekly basis as well as flags any activity that was unusual. The system was clearly documented and handed off to an offshore development team to assist with maintenance and feature implementation with DataSpartan still having oversight for technically difficult challenges.
