Credit Risk Anomaly Detection
Deutsche Bank, London
Sudden volatility, credit downgrades, litigation and reputational events among counterparties carry direct economic cost, regulatory fines and franchise damage. The bank held a growing volume of relevant data, but translating it into usable knowledge was a linear process constrained by manual effort and by warehousing technology that could not handle the volume or the complexity.
Scope of work
Anomaly detection service built across all asset classes in the bank's portfolio, replacing the alternative of a separate model per desk. It identifies:
- Point anomalies, such as a sudden widening in spread
- Pattern anomalies that develop over a longer window
- Movements mapped to the exposures they affect on the banking and trading books
Natural language engine built bank-wide to digest news, market commentary, internal research and ESG and sustainability data, so each anomaly arrives with the context that explains why it occurred.
Early warning signals produced through sentiment and topic modelling applied across that material, monitoring counterparties and the names they depend on.
Persistent insight store built on a scalable cloud data lake and reusable across teams, holding:
- Sentiment and topic output
- Detected anomalies and their context
- Reputational risk, sustainability and ESG scores
One platform combining machine learning, natural language processing, data operations and cloud, designed to be cheaper and more scalable than the existing warehousing approach.
Outcome and value
- Capability that compounds as data volume grows rather than degrading under it.
- Earlier warning on banking and trading book exposures, with the explanation attached to each signal.
- Reduced exposure to reputational damage and regulatory penalties.
- A leadership position in applying sustainability and ESG scores to credit monitoring.