This new update is triggered by the revised Capital Requirements Regulation (CRR2) and Capital Requirements Directive (CRD5). Both documents were adopted by the Council of the EU and the European Parliament in 2019 as part of the Risk Reduction Measures package. The CRR2 and CRD5 included numerous mandates for the EBA to come up with new or adjusted technical standards and guidelines. These are now covered in three separate consultation papers

1. The first consultation paper¹ describes the update of the IRRBB Guidelines themselves.
2. The second paper² concerns the introduction of a standardized approach (SA) which should be applied when a competent authority deems a bank’s internal model for IRRBB management non-satisfactory. It also introduces a simplified SA for smaller and non-complex institutions.
3. The third consultation paper³ offers updates to the supervisory outlier test (SOT) for the Economic Value of Equity (EVE) and the introduction of an SOT for Net Interest Income (NII). Read our analysis on this consultation paper here »

In this article we focus on the first consultation paper. The update of the IRRBB Guidelines can be split up in three topics and each will be discussed in further detail:

• Additional criteria for the assessment and monitoring of the credit spread risk arising from non-trading book activities (CSRBB)
• The criteria for non-satisfactory IRRBB internal systems
• A general update of the existing IRBBB Guidelines

CSRBB

The 2018 IRRBB Guidelines introduced the obligation for banks to monitor CSRBB. However, the publication did not describe how to do this. In the updated consultation paper the EBA defines the measurement of CSRBB as a separate risk class in more detail. The general governance related aspects such as (management) responsibilities, IT systems and internal reporting framework are separately defined for CSRBB, but are similar to those for IRRBB.

Also similar to IRRBB is that banks must express their risk appetite for CSRBB both from an NII as well as an economic value perspective.

The EBA defines CSRBB as:

“The risk driven by changes of the market price for credit risk, for liquidity and for potentially other characteristics of credit-risky instruments, which is not captured by IRRBB or by expected credit/(jump-to-) default risk. CSRBB captures the risk of an instrument’s changing spread while assuming the same level of creditworthiness, i.e. how the credit spread is moving within a certain rating/PD range.”

EBA

Compared to the previous definition, rating/PD migrations are explicitly excluded from CSRBB. Including idiosyncratic spreads could lead to double counting since these are generally covered by a credit risk framework. However, the guidelines give some flexibility to include idiosyncratic spreads, as long as the results are more conservative than when idiosyncratic spreads are excluded. This is because, based on the Quantitative Impact Study of December 2020 (QIS 2020), banks indicated to find it difficult to separate the idiosyncratic spreads from the credit spread.

Also, the scope of CSRBB has changed from the current IRRBB Guidelines. All assets, liabilities and off-balance-sheet items in the banking book that are sensitive to credit spread changes are within the scope of CSRBB whereas the 2018 IRRBB Guidelines focused only on the asset side. Based on the results of the QIS 2020, the EBA concluded that most of the exposures to CSRBB are debt securities which are accounted for at fair value (via Profit and Loss or Other Comprehensive Income). However, this does not rule out that other assets or liabilities could be exposed to CSRBB. It is stated that banks cannot ex-ante exclude positions from the scope of CSRBB. Any potential exclusion of instruments from the scope of CSRBB must be based on the absence of sensitivity to credit spread risk and appropriately documented. At a minimum, banks must include assets accounted at fair value in their scope.

We believe that the obligation to report CSRBB for all assets accounted for at fair value will be challenging for exposures that do not have quoted market prices. Without a deep liquid market, it will be difficult to establish the credits spread risk (even when idiosyncratic risk is included).

Another possible candidate to be included in the scope of CSRBB is the issued funding on the liability side of the banking book, especially in a NII context. When market spreads increase, this could become harmful when the wholesale funding needs to be rolled over against higher credit spread without being able to increase client interest on the asset side. Similar to IRRBB, the exposure to this risk depends on the repricing gap of the assets and liabilities. In this case, however, swaps cannot be used as hedge.

Other products such as consumer loans, mortgages, and consumer deposits, which are typically accounted for at amortized cost, are less likely to be included. This is also stated by the BCBS standards. The BCBS states that the margin (administrative rate) is under absolute control of the bank and hence not impacted by credit spreads. However, it is unclear whether this is sufficient to rule these products out of scope.

Non-satisfactory IRRBB internal systems

The EBA is mandated to specify the criteria for determining an IRRBB internal system as non-satisfactory. The EBA has identified specific items for this that should be considered. At a minimum, banks should have implemented their internal system in compliance with the IRRBB Guidelines, taking into account the principle of proportionality. More specifically:

• Such a system must cover all material interest rate risk components (gap risk, basis risk, option risk).
• The system should capture all material risks for significant assets, liabilities and off-balance sheet type instruments (e.g. non-maturing deposits, loans, and options).
• All estimated parameters must be sufficiently back tested and reviewed, considering the nature, scale and complexity of the bank.
• The internal system must comply with the model governance and the minimum required validation, review and control of IRRBB exposures as detailed by the IRRBB guidelines.
• Competent authorities may require banks to use the standardized approach³ if the internal systems are deemed non-satisfactory.

General update of existing IRRBB Guidelines

Major parts of the guidelines for managing IRRBB have not changed. In the section on IRRBB stress testing, however, a new article (#103) for products with significant repricing restrictions (e.g. an explicit floor on non-maturing deposits – NMDs) is introduced. As part of their stress testing, banks should consider the impact when these products are replaced with contracts with similar characteristics, even under a run-off assumption. The exact intention of this article is unclear. For NII-purposes it is common practice to roll over products with similar characteristics (or use another balance sheet development assumption). Our interpretation of this article is that banks are expected to measure the risk of continued repricing restrictions in an economic value perspective when the maturity of those funding sources is smaller than the maturity of the asset portfolio. This may for example materialize when banks roll over NMDs that are subject to a legally imposed floor.

Another update is the restriction on the maximum weighted average repricing maturity of five years for NMDs. This cap was prescribed for the EVE SOT and is now included for the internal measurement of IRRBB. We believe that the impact of this will be limited since only a few banks will have separate NMD models for internal measurement and the SOT.

Finally, some minor additions have been included in the guidelines. For example, the guidelines emphasize multiple times that when diversification assumptions are used for the measurement of IRRBB, these must be appropriately stressed and validated.

Conclusion

It is expected that the final guidelines will not deviate significantly from the consultation paper. Banks can therefore start preparing for these new expectations. For the measurement of IRRBB, limited changes are introduced in the consultation. Although the exact intention of the EBA is unclear to us, it is interesting to notice that the updated IRRBB Guidelines include the expectation that banks pay special attention in their stress tests to products with significant repricing restrictions. Furthermore, banks must invest in their CSRBB measurement. For their entire banking book, banks need to assess whether market wide credit spread changes will have an impact on an NII and/or economic value perspective. The scope of CSRBB measurement may need to be extended to include the funding issued by the bank. And to conclude, the obligation to measure CSRBB for fair value assets that do not have quoted market prices will be a challenge for banks.

References

1. CP Draft GL on IRRBB and CSRBB.pdf
2. CP Draft RTS on SA.pdf
3. CP Draft RTS on SOTs.pdf

The current version of the IRRBB Guidelines, published in 2018, came into force on 30 June 2019. At that time, the IRRBB Guidelines were aligned with the Standards on interest rate risk in the banking book, published by the Basel Committee on Banking Supervision (in short, the BCBS Standards) in April 2016.

The new updates are triggered by the revised Capital Requirements Regulation (CRR2) and Capital Requirements Directive (CRD5). Both documents were adopted by the Council of the EU and the European Parliament in 2019 as part of the Risk Reduction Measures package. The CRR2 and CRD5 included numerous mandates for the EBA to come up with new or adjusted technical standards and guidelines. These are now covered in three separate consultation papers:

1. The first consultation paper_¹ describes an update of the IRRBB Guidelines themselves. The main changes are the specification of criteria to identify “non-satisfactory internal models for IRRBB management” and the specification of criteria to assess and monitor Credit Spread Risk in the Banking Book (CSRBB). Read our analysis on this consultation paper here »
2. The second paper² concerns the introduction of a standardized approach (SA) which should be used when a competent authority deems a bank’s internal model for IRRBB management non-satisfactory. It also introduces a Simplified SA for smaller and non-complex institutions.
3. The third consultation paper³ offers updates to the supervisory outlier test (SOT) for the Economic Value of Equity (EVE) and the introduction of an SOT for Net Interest Income (NII).

Please note that we recently also published an article about the new disclosure requirements for IRRBB which is closely related to this topic.

Changes to the supervisory outlier test

Banks have been subject to an SOT already since the 2006 IRRBB Guidelines. The SOT is an important tool for supervisors to perform peer reviews and to compare IRRBB exposures between banks. The SOT measures how the EVE responds to an instantaneous parallel (up and down) yield curve shift of 200 basis points. Changes in EVE that exceed 20% of the institution’s own funds will trigger supervisory discussions and may lead to additional capital requirements.

Some changes to the SOT were included in the 2018 update of the IRRBB Guidelines. Next to further guidance on its calculation, the existing SOT was complemented with an additional SOT. The additional SOT was based on the same metric and guidelines, but the scenarios applied were the six standard interest rate scenarios introduced in the BCBS Standards. Also, a threshold of 15% compared to Tier 1 capital was applied. In the 2018 IRRBB Guidelines, the additional SOT was considered an ‘early warning signal’ only.

The new update of the IRRBB Guidelines includes two important SOT-related changes, which are incorporated through amendments to Article 98 (5) of the CRD: the replacement of the 20% SOT for EVE and the introduction of the SOT for NII. Both changes are discussed in more detail below.

Changes to the supervisory outlier test

Banks have been subject to an SOT already since the 2006 IRRBB Guidelines. The SOT is an important tool for supervisors to perform peer reviews and to compare IRRBB exposures between banks. The SOT measures how the EVE responds to an instantaneous parallel (up and down) yield curve shift of 200 basis points. Changes in EVE that exceed 20% of the institution’s own funds will trigger supervisory discussions and may lead to additional capital requirements.

Some changes to the SOT were included in the 2018 update of the IRRBB Guidelines. Next to further guidance on its calculation, the existing SOT was complemented with an additional SOT. The additional SOT was based on the same metric and guidelines, but the scenarios applied were the six standard interest rate scenarios introduced in the BCBS Standards. Also, a threshold of 15% compared to Tier 1 capital was applied. In the 2018 IRRBB Guidelines, the additional SOT was considered an ‘early warning signal’ only.

The new update of the IRRBB Guidelines includes two important SOT-related changes, which are incorporated through amendments to Article 98 (5) of the CRD: the replacement of the 20% SOT for EVE and the introduction of the SOT for NII. Both changes are discussed in more detail below.

Replacement of the 20% SOT for EVE

The first part of the amended Article 98 (5) concerns the replacement of the original 20% SOT by the 15% SOT. While many banks are probably already targeting levels below 15%, we expect that this change will limit the maneuvering capabilities of banks as they will likely choose to implement a management buffer. Note that not only the threshold is lower (15% instead of 20%), but also the denominator (Tier 1 capital instead of own funds). Furthermore, the worst outcome of all six supervisory scenarios should be used, as opposed to worst outcome of just the two parallel ones. Combined, this leads to a significant reduction in the EVE risk to which a bank may be exposed.

Some other noteworthy updates to the SOT for EVE that are not directly related to the CRD amendment are listed below:

• The post-shock interest floor decreases from -100 to -150 basis points and it increases to 0% over a 50-year instead of a 20-year period.
• In the calibration of the interest rate shocks for currencies for which the shocks have not been prescribed, the most recent 16-year period should be used (instead of the 2000-2015 period which is still underlying the shocks for the other currencies).
• When aggregating the results over currencies, some additional offsetting (80% as opposed to 50%) is granted in case of Exchange Rate Mechanism (ERM) II currencies with a formally agreed fluctuation band narrower than the standard band of +/-15%. Currently, only positions in the Denmark Krona (DKK) qualify for this treatment.

Introduction of the SOT for NII

The second part of the amended Article 98 (5) concerns the introduction of an entirely new SOT. It is aimed at measuring the potential decline in NII for two standard interest rate shock scenarios. Compared to the SOT for EVE, the SOT for NII requires many more modeling assumptions, in particular to determine the expected balance sheet development. The consultation paper provides clarity on the approach the EBA wants to take but two decisions are explicitly consulted.

The SOT for NII compares the NII for a baseline scenario with the NII in a shocked scenario over a one-year horizon. The two shocks that need to be applied are the two instantaneous parallel shocks that are also used in the SOT for EVE. Furthermore, the same requirements that are specified for the SOT for EVE apply, for example the use of the floor and the aggregation approach. The two exceptions are the requirements to use a constant balance sheet assumption (as opposed to a run-off balance sheet) and to include commercial margins and other spread components in the calculations. The commercial margins of new instruments should equal the prevailing levels (as opposed to historical ones).

The two decisions for which the EBA is seeking input are:

The scope/definition of NII
In its narrowest definition, the SOT will focus on the difference between interest income and interest expenses. The EBA, however, also considers using a broader definition where the effect of market value changes of instruments accounted for at Fair Value (∆FV) is added, and possibly also interest rate sensitive fees and commissions.

The definition of the SOT’s threshold
Article 98 (5) requires the EBA to specify what is considered a ‘large decline’ in NII, in which case the competent authorities are entitled to exercise their supervisory powers. This first requires a metric. The EBA is consulting two:

• The first metric is calculating the change in NII (the difference between the shocked and baseline NII) relative to the Tier 1 Capital:
• The second metric is calculating the change in NII relative to the baseline scenario, after correcting for administrative expenses that can be allocated to NII:
• where α is the historical share of NII relative to the operating income as reported based on FINREP input.

The pros of the narrow definition of NII are improved comparability and ease of computation, where the main pro of the broader definition is that it achieves a more comprehensive picture, which is also more in line with the EBA IRRBB Guidelines. With respect to the metrics, the first (capital-based) metric is the simplest and it is comparable to the approach taken for the SOT for EVE. The second metric is close to a P&L-based metric and the EBA argues that its main advantage is that “it takes into account both the business model and cost structure of a bank in the assessment of the continuity of the business operations”. It does involve, however, the application of some assumptions on determining the α parameter.

Thresholds for the four possible combinations

For each of the four possible combinations (definition of NII and specification of the metric), the EBA has determined, using data from the December 2020 Quantitative Impact Study (QIS), what the corresponding thresholds should be. Their starting point has been to make the SOT for NII as stringent as the SOT for EVE. Effectively, they reverse engineered the threshold to achieve a similar number of outliers under both measures. We expect that the proposed threshold for any of the four possible combinations will not be constraining for the majority of banks. The resulting proposed thresholds are included in the table below:

Table 1 – Comparison of the proposed thresholds for each combination of metric and scope

The impact of including Fair Value changes seems arbitrary as it increases the threshold for the capital-based metric and decreases the threshold for the P&L-based metric. Also, from a comparability and computational perspective, the narrow definition of NII may be preferred. Furthermore, the capital-based metric is less intuitive for NII than it is for EVE, and consequently, the P&L-based one may be preferred. It is also noted in the consultation paper that if the shocked NII after the correction for administrative expenses (the numerator) is negative, it will also be considered an outlier.

Conclusion

In the past years, many banks have invested heavily in their IRRBB framework following the 2018 update of the IRRBB Guidelines. Once again, an investment is required. Even though there are not many surprises in the proposed updates related to the SOTs, small and large banks alike will need to carefully assess how the changes to the existing SOT and the introduction of the new SOT will impact their interest rate risk management. Banks still have the opportunity to respond to all three consultation papers until 4 April 2022.

References

1. CP Draft GL on IRRBB and CSRBB.pdf
2. CP Draft RTS on SA.pdf
3. CP Draft RTS on SOTs.pdf

ML models

ML models study a dataset and use the knowledge gained to make predictions for other datapoints. An ML model consists of an ML algorithm and one or more hyperparameters. ML algorithms study a dataset to make predictions, where hyperparameters determine the settings of the ML algorithm. The studying of a dataset is known as the training of the ML algorithm. Most ML algorithms have hyperparameters that need to be set by the user prior to the training. The trained algorithm, together with the calibrated set of hyperparameters, form the ML model.

ML models have different forms and shapes, and even more purposes. For selecting an appropriate ML model, a deeper understanding of the various types of ML that are available and how they work is required. Three types of ML can be distinguished:

• Supervised learning.
• Unsupervised learning.
• Semi-supervised learning.

The main difference between these types is the data that is required and the purpose of the model. The data that is fed into an ML model is split into two categories: the features (independent variables) and the labels/targets (dependent variables, for example, to predict a person’s height – label/target – it could be useful to look at the features: age, sex, and weight). Some types of machine learning models need both as an input, while others only require features. Each of the three types of machine learning is shortly introduced below.

Supervised learning

Supervised learning is the training of an ML algorithm on a dataset where both the features and the labels are available. The ML algorithm uses the features and the labels as an input to map the connection between features and labels. When the model is trained, labels can be generated by the model by only providing the features. A mapping function is used to provide the label belonging to the features. The performance of the model is assessed by comparing the label that the model provides with the actual label.

Unsupervised learning

In unsupervised learning there is no dependent variable (or label) in the dataset. Unsupervised ML algorithms search for patterns within a dataset. The algorithm links certain observations to others by looking at similar features. This makes an unsupervised learning algorithm suitable for, among other tasks, clustering (i.e. the task of dividing a dataset into subsets). This is done in such a manner that an observation within a group is more like other observations within the subset than an observation that is not in the same group. A disadvantage of unsupervised learning is that the model is (often) a black box.

Semi-supervised learning

Semi-supervised learning uses a combination of labeled and unlabeled data. It is common that the dataset used for semi-supervised learning consist of mostly unlabeled data. Manually labeling all the data within a dataset can be very time consuming and semi-supervised learning offers a solution for this problem. With semi-supervised learning a small, labeled subset is used to make a better prediction for the complete data set.

The training of a semi-supervised learning algorithm consists of two steps. To label the unlabeled observations from the original dataset, the complete set is first clustered using unsupervised learning. The clusters that are formed are then labeled by the algorithm, based on their originally labeled parts. The resulting fully labeled data set is used to train a supervised ML algorithm. The downside of semi-supervised learning is that it is not certain the labels are 100% correct.

Setting up the model

In most ML implementations, the data gathering, integration and pre-processing usually takes more time than the actual training of the algorithm. It is an iterative process of training a model, evaluating the results, modifying hyperparameters and repeating, rather than just a single process of data preparation and training. After the training is performed and the hyperparameters have been calibrated, the ML model is ready to make predictions.

Machine learning in financial risk management

ML can add value to financial risk management applications, but the type of model should suit the problem and the available data. For some applications, like challenger models, it is not required to completely explain the model you are using. This makes, for example, an unsupervised black box model suitable as a challenger model. In other cases, explainability of model results is a critical condition while choosing an ML model. Here, it might not be suitable to use a black box model.

In the next section we present some examples where ML models can be of added value in financial risk management.

Data quality analysis

All modeling challenges start with data. In line with the ‘garbage in, garbage out’ maxim, if the quality of a dataset is insufficient then an ML model will also not perform well. It is quite common that during the development of an ML model, a lot of time is spent on improving the data quality. As ML algorithms learn directly from the data, the performance of the resulting model will increase if the data quality increases. ML can be used to improve data quality before this data is used for modeling. For example, the data quality can be improved by removing/replacing outliers and replacing missing values with likely alternatives.

An example of insufficient data quality is the presence of large or numerous outliers. An outlier is an observation that significantly deviates from the other observations in the data, which might indicate it is incorrect. Outlier detection can easily be performed by a data scientist for univariate outliers, but multivariate outliers are a lot harder to identify. When outliers have been detected, or if there are missing values in a dataset, it might be useful to substitute some of these outliers or impute for missing values. Popular imputation methods are the mean, median or most frequent methods. Another option is to look for more suitable values; and ML techniques could help to improve the data quality here.

Multiple ML models can be combined to improve data quality. First, an ML model can be used to detect outliers, then another model can be used to impute missing data or substitute outliers by a more likely value. The outlier detection can either be done using clustering algorithms or by specialized outlier detection techniques.

Loan approval

A bank’s core business is lending money to consumers and companies. The biggest risk for a bank is the credit risk that a borrower will not be able to fully repay the borrowed amount. Adequate loan approval can minimize this credit risk. To determine whether a bank should provide a loan, it is important to estimate the probability of default for that new loan application.

Established banks already have an extensive record of loans and defaults at their disposal. Together with contract details, this can form a valuable basis for an ML-based loan approval model. Here, the contract characteristics are the features, and the label is the variable indicating if the consumer/company defaulted or not. The features could be extended with other sources of information regarding the borrower.

Supervised learning algorithms can be used to classify the application of the potential borrower as either approved or rejected, based on their probability of a future default on the loan. One of the suitable ML model types would be classification algorithms, which split the dataset into either the ‘default’ or ‘non-default’ category, based on their features.

Challenger models

When there is already a model in place, it can be helpful to challenge this model. The model in use can be compared to a challenger model to evaluate differences in performance. Furthermore, the challenger model can identify possible effects in the data that are not captured yet in the model in use. Such analysis can be performed as a review of the model in use or before taking the model into production as a part of a model validation.

The aim of a challenger model is to challenge the model in use. As it is usually not feasible to design another sophisticated model, mostly simpler models are selected as challenger model. ML models can be useful to create more advanced challenger models within a relatively limited amount of time.

Challenger models do not necessarily have to be explainable, as they will not be used in practice, but only as a comparison for the model in use. This makes all ML models suitable as challenger models, even black box models such as neural networks.

Segmentation

Segmentation concerns dividing a full data set into subsets based on certain characteristics. These subsets are also referred to as segments. Often segmentation is performed to create a model per segment to better capture the segment’s specific behavior. Creating a model per segment can lower the error of the estimations and increase the overall model accuracy, compared to a single model for all segments combined.

Segmentation can, among other uses, be applied in credit rating models, prepayment models and marketing. For these purposes, segmentation is sometimes based on expert judgement and not on a data-driven model. ML models could help to change this and provide quantitative evidence for a segmentation.

There are two approaches in which ML models can be used to create a data-driven segmentation.  One approach is that observations can be placed into a certain segment with similar observations based on their features, for example by applying a clustering or classification algorithm. Another approach to segment observations is to evaluate the output of a target variable or label. This approach assumes that observations in the same segment have the same kind of behavior regarding this target variable or label.

In the latter approach, creating a segment itself is not the goal, but optimizing the estimation of the target variable or classifying the right label is. For example, all clients in a segment ‘A’ could be modeled by function ‘a’, where clients in segment ‘B’ would be modeled by function ‘b’. Functions ‘a’ and ‘b’ could be regression models based on the features of the individual clients and/or macro variables that give a prediction for the actual target variable.

Credit scoring

Companies and/or debt instruments can receive a credit rating from a credit rating agency. There are a few well-known rating agencies providing these credit ratings, which reflects their assessment of the probability of default of the company or debt instrument. Besides these rating agencies, financial institutions also use internal credit scoring models to determine a credit score. Credit scores also provide an expectation on the creditworthiness of a company, debt instrument or individual.

Supervised ML models are suitable for credit scoring, as the training of the ML model can be done on historical data. For historical data, the label (‘defaulted’ or ‘not defaulted’) can be observed and extensive financial data (the features) is mostly available. Supervised ML models can be used to determine reliable credit scores in a transparent way as an alternative to traditional credit scoring models. Alternatively, credit scoring models based on ML can also act as challenger models for traditional credit scoring models. In this case, explainability is not a key requirement for the selected ML model.

Conclusion

ML can add value to, or replace, models applied in financial risk management. It can be used in many different model types and in many different manners. A few examples have been provided in this article, but there are many more.

ML models learn directly from the data, but there are still some choices to be made by the model user. The user can select the model type and must determine how to calibrate the hyperparameters. There is no ‘one size fits all’ solution to calibrate a ML model. Therefore, ML is sometimes referred to as an art, rather than a science.

When applying ML models, one should always be careful and understand what is happening ‘under the hood’. As with all modeling activities, every method has its pitfalls. Most ML models will come up with a solution, even if it is suboptimal. Common sense is always required when modeling. In the right hands though, ML can be a powerful tool to improve modeling in financial risk management.

Working with ML models has given us valuable insights (see the box below). Every application of ML led to valuable lessons on what to expect from ML models, when to use them and what the pitfalls are.

Machine learning and Zanders

Zanders already encountered several projects and research questions where ML could be applied. In some cases, the use of ML was indeed beneficial; in other cases, traditional models turned out to be the better solution.

During these projects, most time was spent on data collection and data pre-processing. Based on these experiences, an ML based dataset validation tool was developed. In another case, a model was adapted to handle missing data by using an alternative available feature of the observation.

ML was also used to challenge a Zanders internal credit rating model. This resulted in useful insights on potential model improvements. For example, the ML model provided more insight in variable importance and segmentation. These insights are useful for the further development of Zanders’ credit rating models. Besides the insights what could be done better, the ML model also emphasized the advantages of classical models over the ML-based versions. The ML model was not able to provide more sensible ratings than the traditional credit rating model.

In another case, we investigated whether it would be sensible and feasible to use ML for transaction screening and anomaly detection. The outcome of this project once more highlighted that data is key for ML models. The available data was numerous, but of low quality. Therefore, the used ML models were not able to provide a helpful insight into the payments, or to consistently detect divergent payment behavior on a large scale.

Besides the projects where ML was used to deliver a solution, we investigated the explainability of several ML models. During this process we gained knowledge on techniques to provide more insights into otherwise hardly understandable (black box) models.

This exemption has been applied in a diverse way by supervisors and therefore, the treatment of Structural FX risk has been updated in recent regulatory publications. In this article we discuss these publications and market practice around Structural FX risk based on an analysis of the policies applied by the top 25 banks in Europe.

Based on the 1996 amendment to the Capital Accord, banks that apply for the exemption of Structural FX positions can exclude these positions from the Pillar 1 capital requirement for market risk. This exemption was introduced to allow international banks with subsidiaries in currencies different from the reporting currency to employ a strategy to hedge the capital ratio from adverse movements in the FX rate. In principle a bank can apply one of two strategies in managing its FX risk.

1. In the first strategy, the bank aims to stabilize the value of its equity from movements in the FX rate. This strategy requires banks to maintain a matched currency position, which will effectively protect the bank from losses related to FX rate changes. Changes in the FX rate will not impact the equity of a bank with e.g. a consolidated balance sheet in Euro and a matched USD position. The value of the Risk-Weighted Assets (RWAs) is however impacted. As a result, although the overall balance sheet of the bank is protected from FX losses, changes in the EUR/USD exchange rate can have an adverse impact on the capital ratio.
2. In the alternative strategy, the objective of the bank is to protect the capital adequacy ratio from changes in the FX rate. To do so, the bank deliberately maintains a long, open currency position, such that it matches the capital ratio. In this way, both the equity and the RWAs of the bank are impacted in a similar way by changes in the EUR/USD rate, thereby mitigating the impact on the capital ratio. Because an open position is maintained, FX rate changes can result in losses for the bank. Without the exemption of Structural FX positions, the bank would be required to hold a significant amount of capital for these potential losses, effectively turning this strategy irrelevant.

As can also be seen in the exhibit below, the FX scenario that has an adverse impact on the bank differs between both strategies. In strategy 1, an appreciation of the currency will result in a decrease of the capital ratio, while in the second strategy the value of the equity will increase if the currency appreciates. The scenario with an adverse impact on the bank in strategy 2 is when the foreign currency depreciates.

Until now, only limited guidance has been available on e.g. the risk management framework, (number of) currencies that can be in scope of the exemption and the maximum open exposure that can be exempted. As a result, the practical implementation of the Structural FX exemption varies significantly across banks. Recent regulatory publications aim to enhance regulatory guidance to ensure a more standardized application of the exemption.

Regulatory Changes

With the publication of the Fundamental Review of the Trading Book (FRTB) in January 2019, the exemption of Structural FX risk was further clarified. The conditions from the 1996 amendment were complemented to a total of seven conditions related to the policy framework required for FX risk and the maximum and type of exposure that can be in scope of the exemption. Within Europe, this exemption is covered in the Capital Requirements Regulation under article 352(2).

To process the changes introduced in the FRTB and to further strengthen the regulatory guidelines related to Structural FX, the EBA has issued a consultation paper in October 2019. A final version of these guidelines was published in July 2020. The date of application was pushed back one year compared to the consultation paper and is now set for January 2022.

The guidelines introduced by EBA can be split in three main topics:

1. Definition of Structural FX.
   The guidelines provide a definition of positions of a structural nature and positions that are eligible to be exempted from capital. Positions of a structural nature are investments in a subsidiary with a reporting currency different from that of the parent (also referred to as Type A), or positions that are related to the cross-border nature of the institution that are stable over time (Type B). A more elaborate justification is required for Type B positions and the final guidelines include some high-level conditions for this.
2. Management of Structural FX.
   Banks are required to document the appetite, risk management procedures and processes in relation to Structural FX in a policy. Furthermore, the risk appetite should include specific statements on the maximum acceptable loss resulting from the open FX position, on the target sensitivity of the capital ratios and the management action that will be applied when thresholds are crossed. It is moreover clarified that the exemption can in principle only be applied to the five largest structural currency exposures of the bank.
3. Measurement of Structural FX.
   The guidelines include requirements on the type and the size of the positions that can be in scope of the exemption. This includes specific formulas on the calculation of the maximum open position that can be in scope of the exemption and the sensitivity of the capital ratio. In addition, banks will need to report the structural open position, maximum open position, and the sensitivity of the capital ratio, to the regulator on a quarterly basis.

One of the reasons presented by the EBA to publish these additional guidelines is a growing interest in the application of the Structural FX exemption in the market.

Market Practice

To understand the current policy applied by banks, a review of the 2019 annual reports of the top 25 European banks was conducted. Our review shows that almost all banks identify Structural FX as part of their risk identification process and over three quarters of the banks apply a strategy to hedge the CET1 ratio, for which an exemption has been approved by the ECB. While most of the banks apply the exemption for Structural FX, there is a vast difference in practices applied in measurement and disclosure. Only 44% of the banks publish quantitative information on Structural FX risk, ranging from the open currency exposure, 10-day VaR losses, stress losses or Economic Capital allocated.

The guideline that will have a significant impact on Structural FX management within the bigger banks of Europe is the limit to include only the top five open currency positions in the exemption: of the banks that disclose the currencies in scope of the Structural FX position, 60% has more than 5 and up to 20 currencies in scope. Reducing that to a maximum of five will either increase the capital requirements of those banks significantly or require banks to move back to maintaining a matched position for those currencies, which would increase the capital ratio volatility.

Conclusion

The EBA guidelines on Structural FX that will to go live by January 2022 are expected to have quite an impact on the way banks manage their Structural FX exposures. Although the Structural FX policy is well developed in most banks, the measurement and steering of these positions will require significant updates. It will also limit the number of currencies that banks can identify as Structural FX position. This will make it less favourable for international banks to maintain subsidiaries in different currencies, which will increase the cost of capital and/or the capital ratio volatility.

Finally, a topic that is still ambiguous in the guidelines is the treatment of Structural FX in a Pillar 2 or ICAAP context. Currently, 20% of the banks state to include an internal capital charge for open structural FX positions and a similar amount states to not include an internal capital charge. Including such a capital charge, however, is not obvious. Although an open FX position will present FX losses for a bank which would favour an internal capital charge, the appetite related to internal capital and to the sensitivity of the capital ratio can counteract, resulting in the need for undesirable FX hedges.

The new guidelines therefore present clarifications in many areas but will also require banks to rework a large part of their Structural FX policies in the middle of a (COVID-19) crisis period that already presents many challenges.

Alternatives for a calculating compounded SARON

After explaining in the previous article the use of compounded SARON as a term alternative to CHF LIBOR, the Swiss National Working Group (NWG) published several options as to how a compounded SARON could be used as a benchmark in banking products, such as loans or mortgages, and financial instruments (e.g. capital market instruments). Underlying these options is the question of how to best mitigate uncertainty about future cash flows, a factor that is inherent in the compounding approach. In general, it is possible to separate the type of certainty regarding future interest payments in three categories . The market participant has:

• an aversion to variable future interest payments (i.e. payments ex-ante unknown). Buying fixed-rate products is best, where future cash flows are known for all periods from inception. No benchmark is required due to cash flow certainty over the lifetime of the product.
• a preference for floating-rate products, where the next cash flow must be known at the beginning of each period. The option ‘in advance’ is applicable, where cash flow certainty exists for a single period.
• a preference for floating-rate products with interest rate payments only close to the end of the period are tolerated. The option ‘in arrears’ is suitable, where cash flow certainty only close to the end of each period exists.

Based on the Financial Stability Board (FSB) user’s guide, the Swiss NWG recommends considering six different options to calculate a compounded risk-free rate (RFR). Each financial institution should assess these options and is recommended to define an action plan with respect to its product strategy. The compounding options can be segregated into options where the future interest rate payments can be categorized as in arrears, in advance or hybrid. The difference in interest rate payments between ‘in arrears’ and ‘in advance’ conventions will mainly depend on the steepness of the yield curve. The naming of the compounding options can be slightly different among countries, but the technique behind those is generally the same. For more information regarding the available options, see Figure 1.

Moreover, for each compounding technique, an example calculation of the 1-month compounded SARON is provided. In this example, the start date is set to 1 February 2019 (shown as today in Figure 1) and the payment date is 1 March 2019. Appendix I provides details on the example calculations.

Figure 1: Overview of alternative techniques for calculating compounded SARON. Source: Financial Stability Board (2019).

0) Plain (in arrears): The observation period is identical to the interest period. The notional is paid at the start of the period and repaid on the last day of the contract period together with the last interest payment. Moreover, a Plain (in arrears) structure reflects the movement in interest rates over the full interest period and the payment is made on the day that it would naturally be due. On the other hand, given publication timing for most RFRs (T+1), the requiring payment is on the same day (T+1) that the final payment amount is known (T+1). An exception is SARON, as SARON is published one business day (T+0) before the overnight loan is repaid (T+1).

Example: the 1-month compounded SARON based on the Plain technique is like the example explained in the previous article, but has a different start date (1 February 2019). The resulting 1-month compounded SARON is equal to -0.7340% and it is known one day before the payment date (i.e. known on 28 February 2019).

1) Payment Delay (in arrears): Interest rate payments are delayed by X days after the end of an interest period. The idea is to provide more time for operational cash flow management. If X is set to 2 days, the cash flow of the loan matches the cash flow of most OIS swaps. This allows perfect hedging of the loan. On the other hand, the payment in the last period is due after the payback of the notional, which leads to a mismatch of cash flows and a potential increase in credit risk.

Example: the 1-month compounded SARON is equal to -0.7340% and like the one calculated using the Plain (in arrears) technique. The only difference is that the payment date shifts by X days, from 1 March 2019 to e.g. 4 March 2019. In this case X is equal to 3 days.

2) Lockout Period (in arrears): The RFR is no longer updated, i.e. frozen, for X days prior to the end of an interest rate period (lockout period). During this period, the RFR on the day prior to the start of the lockout is applied for the remaining days of the interest period. This technique is used for SOFR-based Floating Rate Notes (FRNs), where a lockout period of 2-5 days is mostly used in SOFR FRNs. Nevertheless, the calculation of the interest rate might be considered less transparent for clients and more complex for product providers to be implemented. It also results in interest rate risk that is difficult to hedge due to potential changes in the RFR during the lockout period. The longer the lockout period, the more difficult interest rate risk can be hedged during the lockout period.

Example: the 1-month compounded SARON with a lockout period equal to 3 days (i.e. X equals 3 days) is equal to -0.7337% and known 3 days in advance of the payment date.

3) Lookback (in arrears): The observation period for the interest rate calculation starts and ends X days prior to the interest period. Therefore, the interest payments can be calculated prior to the end of the interest period. This technique is predominately used for SONIA-based FRNs with a delay period of X equal to 5 days. An increase in interest rate risk due to changes in yield curve is observed over the lifetime of the product. This is expected to make it more difficult to hedge interest rate risk.

Example: assuming X is equal to 3 days, the 1-month compounded SARON would start in advance, on January 29, 2019 (i.e. today minus 3 days). This technique results in a compounded 1-month SARON equal to -0.7335%, known on 25 February 2019 and payable on 1 March 2019.

4) Last Reset (in advance): Interest payments are based on compounded RFR of the previous period. It is possible to ensure that the present value is equivalent to the Plain (in arrears) case, thanks to a constant mark-up added to the compounded RFR. The mark-up compensates the effects of the period shift over the full life of the product and can be priced by the OIS curve. In case of a decreasing yield curve, the mark-up would be negative. With this technique, the product is more complex, but the interest payments are known at the start of the interest period, as a LIBOR-based product. For this reason, the mark-up can be perceived as the price that a borrower is willing to pay due to the preference to know the next payment in advance.

Example: the interest rate payment on 1 March 2019 is already known at the start date and equal to -0.7328% (without mark-up).

5) Last Recent (in advance): A single RFR or a compounded RFR for a short number of days (e.g. 5 days) is applied for the entire interest period. Given the short observation period, the interest payment is already known in advance at the start of each interest period and due on the last day of that period. As a consequence, the volatility of a single RFR is higher than a compounded RFR. Therefore, interest rate risk cannot be properly hedged with currently existing derivatives instruments.

Example: a 5-day average is used to calculate the compounded SARON in advance. On the start date, the compounded SARON is equal to -0.7339% (known in advance) that will be paid on 1 March 2020.

6) Interest Rollover (hybrid): This technique combines a first payment (installment payment) known at the beginning of the interest rate period with an adjustment payment known at the end of the period. Like Last Recent (in advance), a single RFR or a compounded RFR for a short number of days is fixed for the whole interest period (installment payment known at the beginning). At the end of the period, an adjustment payment is calculated from the differential between the installment payment and the compounded RFR realized during the interest period. This adjustment payment is paid (by either party) at the end of the interest period (or a few days later) or rolled over into the payment for the next interest period. In short, part of the interest payment is known already at the start of the period. Early termination of contracts becomes more complex and a compensation mechanism is needed.

Example: similar to Last Recent (in advance), a 5-day compounded SARON can be considered as installment payment before the starting date. On the starting date, the 5-day compounded SARON rate is equal to -0.7339% and is known to be paid on 1 March 2019 (payment date). On the payment date, an adjustment payment is calculated as the delta between the realized 1-month compounded SARON, equal to -0.7340% based on Plain (in arrears), and -0.7339%.

There is a trade-off between knowing the cash flows in advance and the desire for a payment structure that is fully hedgeable against realized interest rate risk. Instruments in the derivatives market currently use ‘in arrears’ payment structures. As a result, the more the option used for a cash product deviates from ‘in arrears’, the less efficient the hedge for such a cash product will be. In order to use one or more of these options for cash products, operational cash management (infrastructure) systems need to be updated. For more details about the calculation of the compounded SARON using the alternative techniques, please refer to Table 1 and Table 2 in the Appendix I. The compounding formula used in the calculation is explained in the previous article.

Overall, market participants are recommended to consider and assess all the options above. Moreover, the financial institutions should individually define action plans with respect to their own product strategies.

Conclusions

The transition from IBOR to alternative reference rates affects all financial institutions from a wide operational perspective, including how products are created. Existing LIBOR-based cash products need to be replaced with SARON-based products as the mortgages contract. In the next installment, IBOR Reform in Switzerland – Part III, the latest information from the Swiss National Working Group (NWG) and market developments on the compounded SARON will be explained in more detail.

Appendix I – II

Contact

For more information about the challenges and latest developments on SARON, please contact Martijn Wycisk or Davide Mastromarco of Zanders’ Swiss office: +41 44 577 70 10.

The other articles on this subject: 

Transition from CHF LIBOR to SARON, IBOR Reform in Switzerland, Part I
Compounded SARON and Swiss Market Development, IBOR Reform in Switzerland, Part III
Fallback provisions as safety net, IBOR Reform in Switzerland, Part IV

References

1. Mastromarco, D. Transition from CHF LIBOR to SARON, IBOR Reform in Switzerland – Part I. February 2020.
2. National Working Group on Swiss Franc Reference Rates. Discussion paper on SARON Floating Rate Notes. July 2019.
3. National Working Group on Swiss Franc Reference Rates. Executive summary of the 12 November 2019 meeting of the National Working Group on Swiss Franc Reference Rates. Press release November 2019.
4. National Working Group on Swiss Franc Reference Rates. Starter pack: LIBOR transition in Switzerland. December 2019.
5. Financial Stability Board (FSB). Overnight Risk-Free Rates: A User’s Guide. June 2019.
6. ISDA. Supplement number 60 to the 2006 ISDA Definitions. October 2019.
7. ISDA. Interbank Offered Rate (IBOR) Fallbacks for 2006 ISDA Definitions. December 2019.
8. National Working Group on Swiss Franc Reference Rates. Executive summary of the 7 May 2020 meeting of the National Working Group on Swiss Franc Reference Rates. Press release May 2020

A replicating framework enables a bank to base its interest rate risk measurement and management on investments with a fixed maturity and price – while the deposits have no contractual maturity or price. In addition, a bank can use the framework to transfer the interest rate risk from the business lines to the central treasury, by turning the investments into contractual obligations. There are two commonly used methodologies for constructing the replicating portfolios: the marginal investment strategy and the portfolio investment strategy. These strategies have the same objective, but have different effects on margin and interest-rate term, given certain scenarios.

Strategies defined

An investment strategy determines the monthly allocation of the investable volume across various maturities. The investable volume in month t ( It ) consists of two parts:

The first part is equal to the decrease or increase in the volume of savings deposits compared to the previous month. The second part is equal to the total principal of all investments in the investment portfolio maturing in the current month (end date m = t ), Σi,m=t vi,m.

By investing or re-investing the volume of these two parts, the total principal of the investment portfolio will equal the savings volume outstanding at that moment. When an investment is generated, it receives the market interest rate relating to the maturity at that time. The portfolio investment return is determined as the principal weighted average interest rate.

The difference between a marginal investment strategy and a portfolio investment strategy is that in a marginal investment strategy, the volume is invested with a fixed allocation across fixed maturities. In a portfolio strategy, these parameters are flexible, however investments are generated in such a way that the resulting portfolio each month has the same (target) proportional maturity profile. The maturity profile provides the total monthly principal of the currently outstanding investments that will mature in the future.

In the savings modelling framework, the interest rate risk profile of the savings portfolio is estimated and defined as a (proportional) maturity profile. For the portfolio investment strategy, the target maturity profile is set equal to this estimated profile. For the marginal investment strategy, the ‘investment rule’ is derived from the estimated profile using a formula. Under long lasting constant or stable volume of savings deposits, the investment portfolio given the investment rule converges to the estimated profile.

Strategies illustrated

In Figure 1, the difference between the two strategies is graphically illustrated in an example. The example provides the development of replicating portfolios of the two strategies in two consecutive months upon increasing savings volume. The replicating portfolios initially consist of the same investments with original maturities of one month, 12 months and 36 months. To this end, the same investments and corresponding principals mature. The total maturing principal will be reinvested and the increase in savings volume will be invested.

Figure 1: Maturity profiles for the marginal (figure on top) and portfolie (figure below) investment strategies given increasing volume.

Note that if the savings volume would have remained constant, both strategies would have generated the same investments. However, with changing savings volume, the strategies will generate different investments and a different number of investments (3 under the marginal strategy, and 36 under the portfolio strategy).

The interest rate typical maturities and investment returns will therefore differ, even if market interest rates do not change. For the quantitative properties of the strategies, the decision will therefore focus mainly on margin stability and the interest rate typical maturity given changes in volume (and potential simultaneous movements in market interest rates).

Scenario analysis

The quantitative properties of the investment strategies are explained by means of a scenario analysis. The analysis compares the development of the duration, margin and margin stability of both strategies under various savings volume and market interest rate scenarios.

Client interest rate
As part of the simulation of a margin, a client interest rate is modeled. The model consists of a set of sensitivities to market interest rates (M1,t) and moving averages of market interest rates (MA12,t). The sensitivities to the variables show the degree to which the bank has to reflect market movements in its client interest rate, given the profile of its savings clients.

The model chosen for the interest rate for the point in time t (CRt) is as follows:

Up to a certain degree, the model is representative of the savings interest rates offered by (retail) banks.

Investment strategies
The investment rules are formulated so that the target maturity profiles of the two strategies are identical. This maturity profile is then determined so that the same sensitivities to the variables apply as for the client rate model. An overview of the investment strategies is given in Table 1.

The replication process is simulated for 200 successive months in each scenario. The starting point for the investment portfolio under both strategies is the target maturity profile, whereby all investments are priced using a constant historical (normal) yield curve. In each scenario, upward and downward shocks lasting 12 months are applied to the savings volume and the yield curve after 24 months.

Example scenario

The results of an example scenario are presented in order to show the dynamics of both investment strategies. This example scenario is shown in Figure 2. The results in terms of duration and margin are shown in Figure 3.

As one would expect, the duration for the portfolio investment strategy remains the same over the entire simulation. For the marginal investment strategy, we see a sharp decline in the duration during the ‘shock period’ for volume, after which a double wave motion develops on the duration. In short, this is caused by the initial (marginal) allocation during the ‘stress’ and subsequent cycles of reinvesting it.

With an upward volume shock, the margin for the portfolio strategy declines because the increase in savings volume is invested at downward shocked market interest rates. After the shock period, the declining investment return and client rate converge. For the marginal strategy this effect also applies and in addition the duration effects feed into the margin development.

Scenario spectrum
In the scenario analysis the standard deviation of the margin series, also known as the margin volatility, serves as a proxy for margin stability. The results in terms of margin stability for the full range of market interest rate and volume scenarios are summarized in Figure 4.

Figure 4: Margin volatility of marginal (left-hand figure) and portfolio strategy (right-hand figure) for upward (above) and downward (below) volume shocks.

From the figures, it can be seen that the margin of the marginal investment strategy has greater sensitivity to volume and interest rate shocks. Under these scenarios the margin volatility is on average 2.3 times higher, with the factor ranging between 1.5 and 4.5. In general, for both strategies, the margin volatility is greatest under negative interest-rate shocks combined with upward or downward volume shocks.

Replication in practice

The scenario analysis shows that the portfolio strategy has a number of advantages over the marginal strategy. First of all, the maturity profile remains constant at all times and equal to the modeled maturity of the savings deposits. Under the marginal strategy, the interest rate typical maturity can vary from it over long periods, even when there are no changes in market interest environment or behavior in the savings portfolio.

Secondly, the development of the margin is more stable under volume and interest rate shocks. The margin volatility under the marginal investment strategy is actually at least one and a half times higher under the chosen scenarios.

An intuitive process
These benefits might, however, come at the expense of a number of qualitative aspects that may form an important consideration when it comes to implementation. Firstly, the advantage of a constant interest-rate profile for the portfolio strategy, comes at the expense of intuitive combinations of investments. This may be important if these investments form contractual obligations for the transfer of the interest rate risk.

The strategy, namely, requires generating a large number of investments that can even have negative principals in case of a (small) decline of savings volume. Secondly, the shocks in the duration in a marginal strategy might actually be desirable and in line with savings portfolio developments. For example, if due to market or idiosyncratic circumstances there is high inflow of deposit volume, this additional volume may be relatively more interest rate sensitive justifying a shorter duration.

Nevertheless, the example scenario shows that after such a temporary decline a temporary increase will follow for which this justification no longer applies.

The choice

A combination of the two strategies may also be chosen as a compromise solution. This involves the use of a marginal strategy whereby interventions trigger a portfolio strategy at certain times. An intervention policy could be established by means of limits or triggers in the risk governance. Limits can be set for (unjustifiable) deviations from the target duration, whereas interventions can be triggered by material developments in the market or the savings portfolio.

In its choice for the strategy, the bank is well-advised to identify the quantitative and qualitative effects of the strategies. Ultimately, the choice has to be in line with the character of the bank, its savings portfolio and the resulting objective of the process.

1. The profile shown is a summary of the whole maturity profile. In the whole profile, 5.97% of the replicating volume matures in the first month, 2.69% per month in the second to the 12th month, etc.
2. Note that this is a proxy for the duration based on the weighted average maturity of the target maturity profile.

An extended version of this article is published in our Savings Special. Would you like to read it? Please send an e-mail to marketing@zanders.eu.

More articles about ‘The modeling of savings’:

• Why is modeling essential
• Approaches to modeling liquidity maturity
• The freedom of variable savings