Manage credit risk more precisely and make stronger business decisions
It’s a new frontier for commercial credit risk management – as the advancement in applied statistical learning techniques and data emerge as a game changer.
Enhanced risk models can improve credit underwriting and help monitor performance, sometimes dramatically. Institutions can build increasingly sophisticated models and algorithms, allowing them to learn from data better and faster, manage credit risk more precisely, and proactively make stronger business decisions.
Today, we are on the cusp of a new breed of credit risk models that incorporate signals from unstructured data, and create better and scalable risk measures to support commercial credit decisions
While these techniques have significant benefits, they come with certain challenges and potential pitfalls, testing cultural readiness. Education and enrollment of senior stakeholders are essential for future investment and implementation. Complexity and opaqueness of these models make it harder for stakeholders, including modelers, validators and users, to grasp the embedded intuition.
To remain competitive, financial institutions often need an arsenal of advanced analytics techniques, strategies and innovative datasets – and a solid approach to lay the groundwork for the path forward.
Our paper focuses on applications of statistical learning in commercial credit, covering wholesale lending, corporate, middle market and small and medium enterprise (SME) segments. We take a deep dive into the power of statistical learning, presenting our observations, research and insights through advising clients, and discussing the benefits that advanced credit analytics offer to financial institutions. Whether you’re a chief credit officer looking to develop or improve existing capabilities, or a financial modeler or data scientist looking to implement an advanced analytics strategy, our paper presents the key information to get you started.
1MAKE BETTER USE OF EXISTING DATA SOURCES:
Such as an institution’s financials, for example, by effectively uncovering nonlinear relationships.
2MAKE USE OF VARIED DATA:
Such as unstructured text data found in news and social media channels.
3INCORPORATE DATA THAT IS UPDATED IN REAL TIME:
Which allows analysis to be timely and avoids stale signals.
Our findings and insights for the path forward
Our findings and insights are based on a combination of Oliver Wyman experience advising clients, and our observations and research in this field.
- We present what banks can learn from experimentation with statistical learning, and provide real examples to help improve underwriting and portfolio monitoring decisions.
- We walk through an exploration of advanced algorithm approaches, explain the causalities, and show how to more effectively extract deeper signals.
- We offer lessons learned, including the challenges and potential pitfalls we have experienced in helping institutions develop the first generation of advanced models.
If your institution is ready, we recommend that you deeply engage your teams with the workings and features of these new models and study the characteristics of calibration data, including the embedded signals and biases. If your institution is hesitating, you might want to consider a pragmatic “middle path” approach which guides institutions toward implementing intelligent changes to traditional techniques without abandoning them. This approach directs developers to understand the structures, transformations, algorithms and the data profoundly to lay the groundwork for a fuller conversion to modern techniques.
We hope that the insights shared help the commercial credit community to advance faster in statistical learning, and offers commercial credit leaders and analysts a practical path to unlock the value for better credit risk management.