We previously discussed Early Warning Systems in the context of black swan events, like the COVID-19 pandemic. With their near real-time capabilities, these systems are reliable allies in countering the negative fallout on financial institutions and their credit risk profiles. Incorporating Model Risk Management best practices into the Risk management of an FI is another effective method to manage and mitigate model risk and calibration risk. As valuation adjustment and managing the materiality of the risk drivers take precedence under the stress caused by unpredictable, potentially catastrophic events, the systematic approach and, technology-aided framework should enable the user and the FIs to effectively manage the risk over this time
After the tumultuous impact of COVID-19 on the global financial sector, countries are bracing up with exit plans that will not only help restore financial stability, but also allow them to be prepared for future black swan occurrences. Whether these collective efforts will buckle or thrive under pressure always remains to be seen. But, in the interim, there are several technology-led interventions that financial institutions can adopt to identify and mitigate model risk and calibration risk in the face of uncertainty.
Many models and approaches are widely used in financial institutions for pricing, valuation, analytics, and risk management of banking book and trading books, which include less quantitative models like scoring models, rating models, default prediction models to highly quantitative models like stochastic volatility models, jump diffusion models and so on. In recent times it has been a standard practice that the materiality of risk drivers and valuation adjustments aligned to an institution’s risk governance framework to deliver desired or optimal results. However, tail events and increase in application of new generation approaches such as Machine Learning algorithms often brings in new uncertainty and less visibility to human oversight, miscalculation or indiscretion, and often find themselves at the centre of heated debates and controversies.
For example, in most of the cases, black swan events such as COVID-19 are not factored in the standard models, as these are tail events, less frequent and very few in number, from a historical perspective. In other words, standard models are based on the principles of central limit theorem, and balanced with a trade-off between flexibility and bias, by ignoring the extreme events. Any attempt to fit the model to extreme events within a limited sample size would result in large residual errors, which cannot be generalized or practically used day-to-day.
The application of new generation models could be new solution to these tail events and we are seeing increase in research activities in this space; however these are still unproven territory or not established practises that required close oversight at user level and at the governance level, which required a systematic framework such as Model Risk Management (MRM) solution. Principles of MRM, aided with technology led interventions, can be used to effectively contain such model risks. We will look at how this can be achieved through a few use cases:
1. Stress testing & what-if scenarios
As we discussed, quantitative models used for routine decision-making do not consider the impact of extreme events. Such models have to be complete with testing for stress/what-if scenarios for strategic decision-making purposes, such as fixing tolerance limits, defining materiality of risk drivers and thresholds etc. These help in ensuring that decisions made on the basis of the models stay within limits, even under extreme circumstances.
Technology can be used to simulate various scenarios and can be used for measuring their impact at various levels.
2. Model risk management framework
A comprehensive MRM Framework would help in identifying and containing various risks arising out of models. A typical framework would include qualitative and quantitative standards, documentation of assumption behind the models, regular audits, validation, issue tracking, developing challenger models, and so on. Models also have to be continuously calibrated and refined by adding/dropping variables based on parameters such as information criteria.
A comprehensive technology-led platform would certainly go a long way in enabling an organisation in rolling out a strong Model Management Framework.
3. Learning from the insurance industry
Largely, in cases of extreme events such as epidemics, the insurance industry, through their Actuarial models – more specifically through a family of models called as ‘Epidemiology Models’ or ‘Compartmental Models’ – tries to predict the financial implication of said events. A graded impact analysis (such as Grade I/II/III severity) is performed on the cash flows. The standard practice in actuarial science is to add economic considerations (such as a GDP drop) to models and design insurance policies based on changes in cash flows, in case of an unfortunate event.
The banking industry can take a cue from this and can perform graded impact analyses of the cash flows based on their portfolios, rather than merely relying on single-point default models. Recently certain epidemiological models have been used in applications for Enterprise Risk Management (ERM). Organisations can incorporate such cross-discipline learning as part of the overall Model Risk Management framework to augment the quantitative models.
Model Risk Management technology solutions effectively enhance the key stakeholder’s ability to identify risk in a timely manner, continuously measure its impact, and manage model risks. In a time that’s laden with uncertainty, it is in the best interests of banks and financial institutions to deploy frameworks that underpin model functionality in unprecedented times, empowering them to withstand shocks, and dismissing possibilities of untimely breakdown.
Views are personal.
The author is CEO, BCT Digital.
