Financial institutions are facing a challenge to deal with high velocity risks; market, credit and liquidity- in environments where batch-based reporting cannot scale against the fast-changing exposures. Recent volatility, such as funding tensions and market run events in 20232024 demonstrates the importance of having more detailed and real-time risk insights. The proposed approach that will be discussed in the paper is a streaming-first architecture of predictive analytics and real-time reporting concerning risk management. Our design incorporates constant data ingestion through market, credit, and treasury sources and a stream processing backbone (Kakfa, Flink/Spark) which allows low-latency feature engineering, identification of anomalies and online inference. We make available a single coherent reference architecture, model portfolio, and evaluation model that strike a balance between the analytic precision of systems and operational security in the form of data lineage, encryption and replicability. Prototyping and case study results illustrate the following important implications: Latency costs can be reduced to sub-second levels; latency can result in measurable improvement in early-warning time over batch baselines; and latency application can result in operationally significant improvements of liquidity control, detection of fraud, or monitoring of credit risk. In addition to the technical results, the design is compatible with regulatory requirements and accuracy, timely reporting, and governance relevant to BCBS 239 and RDARR supervisory expectations.
Pratik Chawande - (Mon,) studied this question.