Key takeaways from US, European & Canadian risk executives’ round tables
US banks: the US roundtable featured 15 participants from 8 GSIBs and 4 top-tier national banks.
European banks: the EU roundtable featured 11 participants from 5 GSIBs and 4 top-tier national banks.
Canadian banks: the Canadian roundtable featured 11 participants from 1 GSIB and 3 top-tier national banks.
For many banks, the third quarter of 2020 is lining up two competing challenges: business planning for 2021, and continuing uncertainty about the duration and magnitude of Covid-19’s impact on their business. As a reference point for risk executives wrestling with these twin challenges, we offer the following industry-wide perspectives from the US, Europe, and Canada on model risk management which we gathered during the past few months.
Model risk management in the post-Covid-19 world: Priorities, challenges, and practical solutions in the US
Covid-19 has brought the need for varying degrees of adjustments across model types. Pricing models have been able to adapt better, since they were tested through the 2008 crisis. On the other hand, credit models require adjustments to account for
- heavy reliance on historical data, and
- abrupt changes in economic data and unemployment claims.
Mortgage prepayment models also have to be adjusted. Consensus has it that commercial credit models are over-estimating losses by a significant margin, while retail credit models are underestimating them.
A need for increased focus on model monitoring was agreed upon by all, especially given that regulators do not provide a clear signal on either frequency of monitoring or performance thresholds. Model monitoring becomes more important during crises, which has caused the frequency of monitoring to become a topic of debate, especially with the explosion of model classes like fraud, marketing, capital markets, ATM, credit models etc. This has also led to an unquestionable need for automation in the model monitoring space. While some banks are high up on the automation curve with almost 100% automated monitoring, many others are on the other end of the spectrum with an almost ‘all-manual mode,’ which calls into question the credibility and reliability of their models.
How is Covid-19 affecting the adoption of automation in MRM in the US?
Overall, the participants recognised a significant need for automation in the MRM space, with special focus on the following areas:
- Annual reviews, monitoring and revalidations, which are high volume and high frequency,
- Model validation documentation, and
- Model inventory – codification of playbooks and procedures.
There are no current vendor systems that offer attractive solutions in all these areas.
With Covid-sensitivity and stress testing workload added to BAU procedures, MRM teams are resource-crunched and feeling this urgent push for automation. This has led to some experimentations in the space, such as the emergence of a futurist ‘smart automation performance and data gathering” concept (model telemetry). This futuristic concept would be capable of transmitting all data from the models, including change in inputs, model runs/usage and recalibrations, and model testing, to a central server. This would help MRM teams with easy access to the live codes and a fully updated inventory of models/lifecycle status. While new models can be built into this ecosystem, legacy ones would have to be retrofitted with a python wrapper to transmit performance data.
How effectively are models being tested in Europe for fitness for purpose, including unintended biases and the post Covid-19 world?
Testing for fitness recognises that each model is a framework, not just a formula. Some firms already conduct thorough model fitness reviews, but discussions get complicated by politics and there’s no defined methodology. More executive backing is needed to make this a standard practice. Continuous model monitoring brings the opportunity of ongoing fitness assessment, which is a significant improvement over fitness assessment just at the time of model approval or re-validation. Overall, progress is slow when so many legacy models are still in everyday use.
During the pandemic, MRM has had to be flexible, pro-active, and prioritise its resources to accommodate adjustments to models disrupted by extreme volatility and/or by drivers affected by the pandemic. Complex models have proved to be the most problematic. Integrating MRM into each model’s lifecycle is a widely shared objective, but still far from reality.
How is Covid-19 affecting the adoption of automation in MRM in Europe?
Automation in MRM which was already underway is being accelerated. A few banks have ramped up automation across the board, including in MRM. Model automation offers big advantages but increases risk management concerns in areas such as ethics, privacy, cyber, and 3rd party risk. Remote working for MRM teams has both helped and hindered BAU and new initiatives. The natural evolution is automated model monitoring, which is now a priority and readily implementable for pricing models. However, for other models, automated monitoring is still in its early stages. The way forward is clearly the integration of automated monitoring into model development. In the meantime, expectations of post-pandemic automation-driven MRM cost reductions seem optimistic.
MRM in the post-Covid-19 world: Priorities, challenges, and practical solutions in Canada
Every kind of model has been affected to some extent. Guard-rails, caps and floors are being applied as needed, depending on the type of model. These overlays are being done in a timely fashion in collaboration with model users, but they introduce increased bias. Model performance monitoring, which normally occurs quarterly, is being conducted more frequently where data timeliness allows.
Market risk models don’t easily accommodate negative interest rates or negative commodity prices. The distorting effect of current price anomalies will persist for a long time, especially in time-series, which automatically adds the newest and drops the oldest price. The debate over stochastic vs, deterministic models has intensified.
Commercial credit models have been problematic because of spikes and swings in unemployment rates, especially models which use quarter-on-quarter change.
By contrast, consumer credit models require negative adjustments to compensate for hardship relief. Credit card applications have dropped off significantly, especially from low-risk borrowers, which may cause skew in model results. It’s challenging to decide whether to apply model overlays or design changes while relief payments and deferrals are going on. There’s no road map for predicting defaults in a W-shaped recovery.
AML models show a high volume of alerts, even though the amount of cash transactions has dropped off, requiring parameter adjustments to filter out false positives. Conversely, AML models aren’t able to flag the widespread fraud in relief programmes.
How is Covid-19 affecting the adoption of automation in MRM in Canada?
Regulators are promoting automation, which is growing primarily in areas where it saves costs, such as model inventory management, challenger models, and standardisation of coding. It’s not clear if automation underway before the pandemic has accelerated. However, automation has improved some model performance; for example machine learning for PPNR forecasting sub-models with high predictive power.
For any questions or details regarding the round table topics, please contact us at: MRMsolutions@evalueserve.com
Join Evalueserve at RiskMinds International!
Evalueserve is a leading analytics partner to Fortune500 companies. Powered by mind+machine™, Evalueserve combines insights emerging from data and research with the efficiency of digital tools and platforms to design impactful solutions. A global team of 4,000+ experts collaborates with clients across 15+ industries.
We use process re-engineering and automation to transform the way you manage risk.
Our services cover MRM, Risk Data and Analytics, Risk Managed Applications. Our tools support delivery of ongoing operations and automate several regulatory requirements (e.g., FRY15, FR 9YC, SCCL, SR 11/10, SR 11/7, BCBS 239) with the use of a centralized data warehouse and reporting tool infrastructure.
Our clients benefit from high quality domain expertise, a tailored approach to suit each bank’s unique infrastructure, and project management located in key financial centers across the US, UK and Europe; lowcost highly skilled teams supporting from delivery centers in Chile, Romania and India.
We number 60 of the world’s leading financial institutions among our valued clients.