A model is a simplified representation of real-world relationships among observed characteristics of a given phenomenon and its underlying driving factors. When applied to banks, it consists of a process transforming a set of inputs into estimates, informing economic decisions.
Technological progress and economic innovation have paved the way for a wider use of data and modelling for economic decision-making. As a result, model management is taking a bigger role in informing and steering processes within financial institutions. This trend will likely continue given the dramatic economic changes which have impacted the global financial system in recent years.
In fact, the assumption of inferring future dynamics from past experience is undermined by dramatic environmental changes. The war between Russia and Ukraine, or the Covid-19 pandemic, represent two of these rare events, and have required the integration of non-traditional and traditional modelling techniques. In this regard, model risk management plays a crucial role in designing model risk appetite, monitoring performances, and assessing uncertainty around outcomes. The impacts of disruptive events forced banks to analyse models’ capabilities under extreme scenarios. At the same time, these events have been a catalyst for financial institutions to think differently, inducing them to shift towards ranges of potential outcomes, ideally based on a given distribution, instead of relying only on punctual measures.
Big challenges, big questions, big improvement margins
By developing a model, banks commit to a relevant effort in terms of data collection, methodological design, communication among various stakeholders, and IT infrastructure, among others. Each of these aspects is crucial to ensuring consistency and effective usability of model outcomes. In this regard, there is a worthwhile difference between silo models and complex frameworks (in effect, based on a combination of interacting models).
Recently, much attention has been devoted to checking models’ statistical properties and validating them against development samples. Conversely, very little attention has been paid to scenarios beyond these models being developed and calibrated. Extensive ‘what-if’ analyses should inform the job of model risk managers on a day-by-day basis. At the same time, the key challenge going forward relates to interaction among models.
Think of a balance sheet projection for stress-testing purposes: dozens of models and hundreds of parameters interact with each other. Are we comfortable with the integrity of the entire model chain? What happens if a model does not work properly? How great is its impact on critical key performance indicators (KPIs)? Model risk managers are asked to provide comprehensive answers to all these questions.
Improving governance procedures and accountability
In other words, complex integrated schemes now inform banking processes. Institutions are therefore required to effectively map these models on a standalone basis, as well as components of integrated frameworks. Ongoing status assessment is then required in light of a well-defined model risk appetite design. Clear tolerance limits are useful guidelines for model usage.
Therefore, recalibration, refactoring, model enhancement and rebuilding are part of the comprehensive process in which model risk appetite operates as a guideline, informing the dialogue between various actors across the modelling chain.
Model risk management is the main actor to foster model resilience, promote continuous innovation and optimize modelling processes.
Climate risk modelling
Climate risk is part of this evolving ecosystem and certainly contributes to enriching the spectrum of applications. There is a lot of common features with balance sheet stress-testing, financial and strategic planning. In all of them, linkages across various models are fundamental.
Macroeconomic scenarios feed satellite models. The latter are then used as a transmission mechanism towards end-use KPIs. A well-structured governance framework based on comprehensive model management tools will accompany banks’ evolution towards the automatisation of all their major processes.
AI and machine learning: future of modelling?
Artificial intelligence (AI) and machine learning (ML) are becoming popular for capital strategies, and risk management assessments. For example, credit risk advanced internal ratings-based (AIRB) models have been validated by regulators given their outstanding performance – both in and out of sample.
Various approaches have been developed to improve their explainability. In all cases, a clear correlation between statistical techniques and economic theory should inform the next generation of models adopted by banks.
Looking at the future, multi-agent reinforcement learning and other similar techniques –extensively used in, for example, autonomous driving and robotics – are likely to be a reference point for common tasks performed by financial institutions on a continuous basis.
To find out more about risk management modelling, register for Prometeia’s forthcoming webinar on April 5, 3 PM CEST, here.