Modelling of Interest-Only Mortgage Portfolios

The large volume of interest-only (IO) mortgages, coupled with a high concentration of maturities in certain years and the specific risks associated with IO loans, is posing a significant risk to the balance sheets of most Dutch banks. As a result, the ECB and DNB have been compelled to intervene. In recent years, the ECB has issued multiple letters to large Dutch banks outlining supervisory expectations on various aspects of interest-only mortgages, including risk management, modelling, and pricing, with the objective of reducing and de-risking these portfolios. More recently, smaller Dutch banks have also received similar guidance from the DNB. One of the main challenges for risk management is accurate risk classification and modelling of the credit risk on IO mortgage portfolios. This is challenging due to the limited availability of historical data on matured loans. This challenge is made even more complex by anticipated changes in risk identification—such as those driven by client outreach programs to collect borrower information—and risk classification, including updates to forbearance policies and unlikeliness-to-pay (UTP) frameworks. This paper aims to support credit risk modellers in navigating these rapidly evolving conditions. It offers specific insights and considerations for modelling interest-only portfolios in light of regulatory expectations and market dynamics.

Interest-only (IO) mortgages have become a focal point of regulatory concern in the Netherlands, reflecting the unique risk profile they present to both lenders and the financial system at large. These mortgages, while providing borrowers with the advantage of lower monthly payments, carry significant uncertainties around long-term repayment behaviour due to the lack of information on repayment source and lack of insight in the financial position of borrowers. Over the past years, the European Central Bank (ECB) has raised supervisory expectations for larger Dutch banks, emphasizing the need for enhanced risk management, improved modelling approaches, and refined pricing strategies. Furthermore, the AFM has intervened and recently the Dutch Central Bank (DNB) has extended similar guidance to smaller banks, signalling a unified effort to de-risk interest-only mortgage portfolios across the sector.

Such regulatory developments have also taken place in other European countries. For example, in England, the PRA issued specific policy updates in 2019-2020 to clarify how firms should treat RIO (Retirement Interest-Only) mortgages in their risk assessments and capital calculations. These RIO mortgages differ from traditional interest-only mortgages as the loan is only repaid upon a life event (such as moving into long-term care or passing away), whereas traditional IO mortgages have a fixed repayment date.

The challenges for credit risk modellers are both technical and strategic. Historical data on matured interest-only loans is sparse, limiting the ability to draw reliable conclusions about repayment patterns. In addition, uncertainty regarding future changes to risk classification frameworks—such as updates to forbearance and unlikeliness-to-pay (UTP) standards—and the uncertain effect on risk parameters further complicate the task of building robust and future-proof credit risk models. Another aspect is the lack of up-to-date information on source of repayment and income and assets of clients to repay

This article serves as a practical guide for credit risk modellers and senior management managing interest-only mortgage portfolios. It explores modelling considerations and strategies tailored to a rapidly evolving environment, where simplicity and adaptability are critical. Building on insights from regulatory expectations and industry best practices, the paper highlights key risk drivers, the role of model overlays, and the importance of leveraging intermediate results from client outreach programs. By offering these perspectives, we aim to equip modellers and senior managers with the tools to navigate evolving conditions and support proactive portfolio management strategies.

Finding specific interest-only risk drivers

ECB is definite that IO mortgages have specific risk characteristics and should be considered as a separate sub-segment. The ECB expects banks to model interest-only (IO) mortgages separately from amortizing mortgages to accurately capture the unique risks associated with IO portfolios. These risks can be captured by applying (sub-)segmentation and implementing specific IO risk drivers within the models.

Defaults due to arrears on periodic payments are generally less common in IO mortgage portfolios compared to amortizing portfolios. The monthly instalments for IO loans are lower, excluding principal repayments. However, after an interest rate reset, IO clients may face a relatively larger increase in monthly instalments than annuity clients, particularly since the outstanding exposure for IO loans is not expected to decrease over time. The absolute monthly instalment amount, any foreseen increases in instalment amount due to interest rate reset, and voluntary prepayment can provide insight on the credit risk of the client.

The relevance of specific IO risk drivers is also closely tied to default and forbearance frameworks unique to IO loans. For instance, indicators predicting IO-specific defaults—such as through the “unlikely to refinance” UTP trigger—can enhance risk capture but depend heavily on the bank’s forbearance and definition of default (DoD) frameworks.

It is also crucial to consider the time horizon when analyzing risk differences between IO and amortizing mortgages. While 12-month default rates may appear similar, cumulative default rates over longer periods could diverge significantly. Additionally, representativeness should be carefully managed when comparing default characteristics. For example, interest rate profiles can vary widely, so fair comparisons should focus on cumulative default rates for loans originated in the same period. Analyzing differences in default rates and distributions for different segments can help identify where IO risk diverges most from amortizing loans, for example in segments with high IO exposure (e.g., loans exceeding €100,000) combined with high LTV. These insights can further substantiate (sub-)segmentation decisions, such as introducing high-risk IO sub-segments or determining appropriate risk drivers for the models.

Such analyses may reveal the need for different segmentation in IFRS 9 models compared to IRB models given the difference in time horizons. Nonetheless, many banks opt for consistent segmentation across both frameworks, as IFRS 9 models often leverage IRB models.

Examples of IO-specific risk drivers that can be assessed include:

• IO-LTV: Loan-to-Value ratio considering only the IO portion of the loan.
• IO-LTI: Loan-to-Income ratio considering only the IO portion of the loan.
• Voluntary prepayment level: The frequency and extent of voluntary repayments on IO loans.
• Time until interest rate reset: Combined with the difference between contractual and market interest rates.
• Mortgage interest deductibility: The extent to which tax benefits influence borrower behaviour.

Moreover, the repayment strategy (i.e. repayment, refinancing, or the sale of collateral) may influence the relevance of specific risk drivers. Additionally, it could serve as a potential criterion for sub-segmentation.

Building future-proof models

When developing a credit risk model, modellers should prioritize building a framework that remains robust and adaptable to changing conditions, ensuring its validity over multiple years.

At model initiation, it is important to identify key policy assumptions. These include those related to origination, collateral management, and risk classification for interest-only mortgages. Some assumptions may remain valid in the medium to long term, while others might need reassessment.

This applies in particular to the definition of default, as it forms the foundation of the entire credit risk model.

In rapidly evolving environments, simpler models that focus on capturing key risk drivers without unnecessary complexity are generally preferable. Such models depend on fewer assumptions, reducing the likelihood of inaccuracies over time as conditions change.

Sensitivity analysis can be employed when assessing the performance of the model to assess the potential impact of the data developments on model output. For instance, assess the impact of improvements in the data quality of LTI. Different scenarios can be simulated to assess the potential impact of data obtained through client outreach on credit risk estimates.

This can also inform the need for periodic recalibration, which can be employed to ensure the model remains current. Particularly when new data becomes available at a rapid pace or if the new data significantly differs from historical trends.

For risks that evolve too quickly to be addressed effectively through recalibration, post-model adjustments, such as model overlays, offer a more practical solution. These adjustments are both more flexible and faster to implement, making them well-suited for dynamic conditions.

Quantifying model overlays

Regulations require banks to periodically update borrower information in order to assess creditworthiness. These assessments are expected to be performed at least every three years. This has not been market practice in the Dutch mortgage sector. In the Netherlands, creditworthiness assessments, including repayment tests, are typically conducted only at the time of loan origination.

For interest-only mortgages, large Dutch banks are expected to implement client outreach programs and to perform periodic in-depth creditworthiness assessments.

The introduction of those periodic assessments will significantly enhance the availability of certain data, particularly income-related variables. Additionally, due to the up-to-date data it will be easier to timely identify clients experiencing financial difficulties. This subsequentially leads to more accurate classification and modelling of risks.

Predicting the extent of the increase in identified Stage 2 and Stage 3 (defaulted) clients during the outreach programs in advance is challenging. Intermediate results from the outreach should be leveraged to inform the quantification of model overlays.

It is essential to address potential biases, such as the possibility that riskier clients may be less likely to participate in the outreach, as well as issues of representativeness, especially if the outreach largely targets high-risk (or low-risk) interest-only clients.

Conclusion

IO mortgage portfolios contain specific risks that require  significant adjustments to credit risk models. Supervisors expect banks to model IO mortgages separately from amortizing mortgages. This can be done by introducing specific IO segments and implementing IO risk drivers.

Effectuating the required model adjustments is challenging due to the limited availability of historical data on matured IO loans, the use of new data elements and the lack of up-to-date borrower information. The latter two will need to be solved by client outreach programs.

The required changes in risk classification, credit risk models and incorporation of new and more up-to-date client data creates the opportunity to build a future-proof credit risk framework with a strong foundation. This is also required by supervisors.

To navigate these opportunities, simple models that focus on capturing essential risk drivers are recommended, avoiding unnecessary complexity that may become inaccurate when conditions change. Sensitivity and scenario analysis can be employed to predict the stability of the model in future conditions. When dealing with changing risks or areas lacking sufficient historical data, post-model adjustments such as overlays offer a short term solution but structural solutions and adaptability remains key priority.

Intermediate results from client outreach programs should play a critical role in quantifying these adjustments, as well as in refining portfolio forecasts. It is essential to remain vigilant about potential biases and representativeness issues to ensure that the insights gained are accurate.

Given that banks are forced by supervisors to address current uncertainties on IO, a clear strategy on risk classification, data collection and modelling is a must have. We are convinced that this will help in navigating through the current uncertainties on IO where simplicity, flexibility and continuous interactions between modelling, portfolio management and policy makers are the cornerstones of the structural solution.

This article is written by Arie van der Plas, Stephan Runderkamp and Lisa Lansu.