3rd Annual
Credit Risk Modeling and Management for Financial Institutions
ENTER YOUR DETAILS FOR THE CONFERENCE AGENDA
3. How can we align AI and ML modelling strategies with long-term business goals?
Model developers and owners typically focus on model performance measures for areas such as model accuracy and sensitivity to changes in the underlying data. A bank model’s ongoing performance is monitored and managed by ensuring model performance indicators are within risk tolerances. Linking model performance to business results requires a second step in model evaluation. For example, bank regulators have acknowledged that AI and ML techniques can improve model predictive accuracy and, by taking new data into account, have the potential to expand access to bank credit. This suggests AI and ML techniques might improve business outcomes in several ways—fewer delinquencies and losses on approved loans, increases in loan volume without sacrificing credit standards, and more rapid and less costly model development cycles. Model owners should seek ways to determine a model’s effect on the organization’s key performance indicators (KPIs) and, if there is no good fit of a model with KPIs, evaluate the model’s effect on net revenues.
1. What would you say is the biggest challenge right now within credit risk modeling?
Market volatility has increased dramatically in recent years due to environmental, political, and international events (e.g., Covid-19 pandemic, supply chain bottlenecks, and the War in Ukraine). For example, the average monthly U.S. civilian unemployment rate between 1950 and 1980 was 5.2 percent and rates remained under 11 percent even during the banking crises of the 1980s, early 1990s, and 2007-2009 U.S. financial crisis. With the onset of the Covid-19 pandemic, however, unemployment rates rose dramatically between March and April 2020, increasing from 4.4 percent to 14.8 percent, respectively. Similarly, there have been substantial increases in inflation, interest rates and office vacancy rates following the pandemic. Consequently, credit risk models for loan default and loss given default going into recent periods were based on data that is much different than what was experienced post-2019 and do not extrapolate well over unseen events and volatile periods. This pattern of market volatility and, importantly, new causes of volatility, make it difficult to model credit risk outcomes.
2. How can we ensure the ongoing refinement and dynamic adaptation of models using ML and AI?
There are four aspects to this question that come to mind—the mathematical algorithms used by the model, the computer programs that implement the algorithms, the data on model features and outcomes used to calibrate the model, and the IT systems used to process the model. While the mathematical algorithms many AI and ML models rely on have been around for decades it was the combination of improved computer processing capability, more affordable and enhanced data storage capacity, increases in the scale and scope of information storage, and freely available open-source computer languages for statistical, AI and ML models that fueled the growth in AI and ML model use by banks and other groups. Along with these developments is the expansion of online computer program repositories that can be used for project development. Keeping abreast of developments in the aforementioned areas and taking full advantage of these developments will help ensure ongoing model refinement and adaptation.
Gain invaluable insights into Credit Risk Management from John O'Keefe, Independent Consultant and Financial Analyst at FDIC, set to speak at the upcoming 3rd Annual Credit Risk Modeling and Management for Financial Institutions Conference, March 18-20, 2024, in New York, NY.
John O’Keefe is an independent consultant advising in the areas of bank model risk management and deposit insurance. Mr. O’Keefe’s work focuses on quantitative analytics for banks and deposit insurers. Recent engagements include developing a liquidity stress test model for the deposit insurer of Ecuador (COSEDE), a Toronto Centre workshop on early warning systems and use of financial soundness indicators for the National Bank of Ukraine, a Toolkit for Developing a Deposit Insurance System, and a Toronto Centre workshop for new and established deposit insurers on insurance scheme design.
March 18-20, 2024
New York, NY
John O'Keefe will be hosting the first half of the post-conference workshop on March 20th.
Workshop: Deliver a sustained competitive advantage by enhancing predictive accuracy and interpretability with AI
- Explore and optimizing model architecture of interpretable model design
- Assess best practices in model parameter setting for improving model robustness and resilience
- Navigate advanced developments from the model development and validation perspectives
- Highlight the importance of ongoing model refinement and dynamic adaptation
- Align AI and ML modelling strategies with long-term business goals
For registration pricing and multiple attendee discounts, please contact:
Ayis Panayi
Digital Marketing & PR Executive
All Rights Reserved. marcus evans ® 2024
An Interview with John O'Keefe, Independent Consultant and Financial Analyst (Quantitative) at FDIC
4. In terms of our format, (as you know our conferences are informal and intimate peer-led meetings where all speakers and delegates are senior executives from top financial institutions), how do you see this assisting you with overcoming the challenges the industry currently faces.
Peer-to-peer exchanges of information provide a broader perspective than the experiences of any one institution can offer. In my work on bank model reviews, I have seen how a wide variety of approaches to risk measurement and management through models for different business franchises—e.g., large industrial loan companies, small niche banks and fintech lenders. This has given me an appreciation for the principle of proportionality in model risk management as well as specialization in model development. Having conducted model reviews since 2011 I have witnessed advances in modelling techniques and capabilities with increasing focus on AI and ML for virtually all business areas—e.g., credit, treasury, BSA/AML, and external fraud detection.
These modelling advances offer rewards that come with costs—loss of model transparency and greater potential for model overfitting. Model developers should keep abreast of the specifics of new techniques and balance these tradeoffs. Finally, the importance of data governance has been heightened by the availability of advanced approaches in modelling bank risks that can take advantage of large volumes of data and new types of data.
