In this article Deloitte's Financial Risk Management team looks at Basel Pillar 2 Economic Capital (‘EC’) modelling in banking with a particular focus on Credit Risk models. In particular, we look at the changing use of this well-seasoned model type, starting with an overview of the resurgence of EC, challenges for both retail and corporate parameterisation, and the Concentration Risk question. We finish by considering the best way to prioritise development effort between model design & parameterisation, how EC can be used and some of our EC expert’s considerations for enhancing an EC framework.

The resurgence of Economic Capital

The upsurge in development of EC models in the last 18 months and their (re-)adoption in core banking processes such as Risk Appetite, Risk-based pricing and limit setting is being driven by both external and internal factors. Increasing Regulatory scrutiny and expectation is clearly a driver, but so is a resurgent awareness of the advantages that such models can bring when used in core strategic business processes such as pricing or capital allocation. Banks are once again realising that EC models can help drive improved returns on risk and thus long term enhanced profitability. For those who either turned off their EC models, or never had them in the first place, this can represent a relatively significant cultural shift, but one many believe they must make if they are to keep up with competitors. Interestingly, this time it is not just large multinational banks leading the charge – they are joined by a battalion of smaller banks who have become vocal advocates for a new “simpler” variant of EC model. Smaller banks do not necessarily want or need the large Monte-Carlo simulation engines of more traditional EC models, championed in the modern era by Moody’s Risk Frontier, but instead are seeking generally correct solutions with an accuracy-complexity trade-off aligned to their size. They are comfortable that, what they lose in modelling accuracy, they more than make up for with a deeper understanding of their core portfolios. Simpler modelling solutions, with correspondingly simpler data requirements, allow cost-efficient, speedier embedding within business processes and ensure the business knows how to use its new toolkit to optimise performance. Smaller banks also often need solutions tailored to a banking segment that has traditionally been under-modelled/parametrised in terms of understanding the underlying correlation structures that drive tail loss events – the Retail segment.

The Retail parametrisation problem

Historically, EC model developers have struggled to calibrate for Retail portfolios. With little appetite to purchase expensive external datasets and only short annual time series of internal data, accurate parameter assessment has often been eschewed in favour of, often very limited, expert-driven estimates. When Retail portfolios make up a material part of a bank’s portfolio such an approach is often seen as a key weakness - one that limits the potential for advanced use of the model’s outcomes. Recent advancements in Retail EC parameterisation have started to offer more credible alternatives to these qualitative assessments, however. These new approaches make use of global experience, techniques like back-casting of default rates and, perhaps most critically, take into account the changing “through-the-cycle” risk profile of the bank’s portfolios, which has been a key source of mis-parameterisation in the past.

Understanding Concentration risk in a global context

Traditionally, EC models designed for international banks have made use of a bank’s internal data - data taken from across the world and used in the context of an already globally diversified organisation – they could thus comfortably make the assumption that losses in one region would be offset by returns in others. As a result, such banks are usually not overly concerned with the specifics of any single region. For smaller banks, sometimes now implementing EC for the first time and often heavily concentrated in one region or country, this is a luxury they cannot afford. On one hand, Regulators are asking for the quantification of the extra Concentration risk capital needed over the “globally correct” Pillar 1 level, itself no simple matter without access to a wide set of global default rates or asset correlations. On the other hand their own region’s history over the last 10 to 15 years is often unlikely to be the most probable or “on average” outcome for that same period. Some North Western European or Asian countries, for instance, have no recent stress experience while, conversely, the “PIIGs” of Europe (Portugal, Italy, Ireland and Greece) have stresses that can be considered extraordinary. These two challenges remain a considerable problem for new order EC modellers as they try to build models that are both internally credible and acceptable to regulators. As with all EC modelling the emphasis must remain on capturing the true risks the bank faces, however, auditable quantitative & qualitative justification of all choices made is key.

Pitfalls in Corporate EC Parametrisation

For corporate portfolios there is a well-established tradition of using corporate equity time series for parametrisation. The danger - one shared with all modelling problems but particularly exaggerated here - lies in the assumption that the solution here is simple and thus “objective”. The reality is that modellers must make decisions around the appropriate equities to map to the bank’s corporate portfolios and how wide they cast their net can have material consequences. Implicit within these modelling choices are statements about how the bank views the world and the inherent risk it faces (is corporate default rate volatility specific to its region or would global levels be reached in a crisis for instance). Allowing modellers to make and implement such statements, rather than the bank’s management, has historically been all too common and can be viewed as one of the root causes of the loss of credibility that EC models suffered following the Global Financial Crisis. Advancements in model risk and governance frameworks have focused on addressing exactly this concern, however, when it comes to steps such as assessing a portfolio’s “asset correlations” (an opaque concept at the best of times), we recommend continued vigilance.

Prioritising between model design and parametrisation

EC model development is typically split into two key components - parameterisation and model design. Developments in the latter have recently focused on simplifying the EC engines of the past (though IFRS 9 related developments have also introduced some interesting new features particularly around PiT vs TTC PD use and performing migration losses). A key challenge for smaller banks who do choose to go down the relatively complex simulation route, even if in a simplified manner, is managing the large computation and storage requirements. On the other hand, if banks choose to avoid using simulation altogether, they must accept a ceiling on the accuracy they can achieve. Often, however, this problem is secondary to the more pressing issue of correct parametrisation. EC Asset Correlations, for instance, which form the bedrock of all Credit Risk EC models, remain poorly understood even by seasoned Credit Risk veterans. Even when Asset Correlations are well understood the wide variety of approaches available for their measurement, as well as a complex portfolio-specific web of pros and cons, mean it is often easy for banks to go down the wrong path.

The many uses of EC

EC models can be used in a variety of banking risk management processes and it is important that the design chosen aligns to those intended uses. The typical uses are:

  • ICAAP and Pillar 2 capital requirements;
  • Risk appetite;
  • Limit setting;
  • Capital allocation;
  • Risk-based pricing;
  • Risk-adjusted performance measurement; and
  • Remuneration

While these uses work via diverse mechanisms and through different business processes they share one general goal (with the possible exception of Risk Appetite which we will come to shortly) – to optimise the bank’s long term performance versus the risk it is taking and in so doing maximise long run average ROE. A challenge to this view is that, with Regulatory Capital (“RC”) often the binding capital constraint, EC model outputs can sometimes be viewed as redundant. This view, however, ignores the many weaknesses in the RC risk assessment and assumes that without RC banks would not hold any capital at all, which is clearly also not right. The question a bank’s management need to answer before they can decide how to use EC is “how much do they wish to minimise true losses under extreme scenarios versus a relatively risk-insensitive RC?”

Leaving aside the question of total capital requirements, EC models also have the capability of informing Senior Management about the true downside P&L volatility of their portfolios which, if calibrated correctly, can be an invaluable tool for effective Risk Appetite control. While some banks do so by looking at extreme tail losses (99.9% Confidence Interval VaRs, etc), assuming that by controlling such measures general portfolio loss volatility will also be limited, others cut out the middle man and re-run their EC models at low Confidence Intervals such as 1 in 5 year or 1 in 10 year events, allowing them to directly target their desired level of downside portfolio P&L loss potential.

What are banks doing now – and what should they do

As banks around the world begin to refocus attention on EC modelling, we recommend the following six stages to building an enhanced EC framework:

  • Define your EC uses, objectives and principles – ensuring exec-level engagement early in the process will help ensure the other stages run smoothly
  • Perform an EC framework review - what do you have in place already that can be leveraged? Define your EC Target State and perform a gap analysis/assessment against your current capabilities.
  • Develop modelling solutions to fill the gaps between your current and target state – be sure to do your homework and ensure the solutions you implement are accurate, industry standard and acceptable to regulators
  • Perform Independent Model Validation on your refreshed EC solution and embed as part of your business-as-usual model maintenance lifecycle
  • Develop a phased embedding strategy – getting ahead of any necessary cultural change implications is key
  • Finalise your model lifecycle including at least annual updates for strategy change, data/portfolio change, industry developments, model validation findings, business feedback and use changes.

If you would like to know more about any of the issues discussed in this paper, please contact Hinnerk Fahrenkamp or William Shields who as members of Deloitte UK’s Financial Risk Measurement Quantitative Analytics team have a combined 25+ years’ experience building and embedding EC models, and would be happy to answer any of your queries. Though based in the UK, they provide support to firms all over the world, with recent EC engagements in Ireland, Australia and Indonesia.