Measuring and managing risks embedded in non-maturiing deposit portfolios is a topic that many banks struggle with. The very nature of non-maturing products gives a lot of uncertainty regarding the expected cash flows and interest rates. However, for many (retail) banks, non-maturing deposits are their most important source of funding. This complexity, combined with the materiality, makes this a very challenging topic. In this article, we will outline the most important modelling concepts, that constitute the market practice in this area. For each modelling concept, we will provide a short outline and some guidance and the appropriate use of these models.

Cash flow model

Changing market circumstances, such as market interest rate and/or liquidity spreads changes, have effect on the client rate banks pay on the on-demand savings and on withdrawals customers make on their deposits. The cash flow model explicitly models these behavioral aspects on savings, the client rate and savings volume, in terms of their explanatory factors.

Banks can use the resulting cash flow profile to derive market values and basis point value (BPV), by summing the discounted savings volume and client rate per tenor. Because the model is able to translate market rate scenarios and liquidity spread scenarios towards client rates and cash flows, banks can use the model for earnings measurement and management, and for stress testing. Depending on a bank’s modeling preference, the effects of liquidity spread risk on client rates can also be measured. The cash flow model consists of two sub-models, one for the client rate and one for savings volume. 

The client rate

On-demand savings rates are expected to track market rates, and typically increase (decrease) when market rates increase (decrease). Figure 1 illustrates a couple of typical relationships between market and client rates:

  • Client rates typically react slowly to market rate changes. The ‘repricing speed’ measures the pace at which changes in the market rates are reflected in the client rates. This can range from a couple of days to multiple months or years.
  • Typically, a market rate shock is only partly passed through to the client rate, as the margin on savings is typically interest rate dependent. The ‘pass-through rate’ measures the extent to which such a market rate shock is passed through.
  • Typically, client rates react in an asymmetric manner to upward versus downward market rate moves. Banks are more likely to lower the client rate following a market rate decrease, than the reverse case.
  • Although not reflected in the figure below, particularly since the outbursts of the financial and Euro crises, some banks have taken into account liquidity spreads when pricing their non-maturing products (NMP) volume. For example, when liquidity spreads (and thereby the debt financing costs) soared, banks aimed to attract more NMP volume by pricing it more attractively for its NMP clients.
Figure 1: Illustrative example of the relation between a reference rate and the client rate. Client rates tend to lag market rates.

Figure 1: Illustrative example of the relation between a reference rate and the client rate. Client rates tend to lag market rates.

Considerations for model estimation process

Banks typically use a linear regression model to forecast the client rate based on the market rates and spreads. A standard approach would be to select a number of explanatory variables and select the model with the combination of explanatory factors that aligns best with expert expectations. Most banks include a number of tenors for swap rates and include moving averages of these rates to incorporate for the typical slow reaction of client rates towards market rate changes. Not all banks include liquidity spreads and the moving averages in the model selection process yet, but it is considered best market practice to do so. In the model selection process, expert expectations on the coefficient signs, pass-through-rates, repricing speeds and stressed scenario performance can be used to select the model of use.

Savings volume

Because there is no contractual maturity on on-demand savings products and customers can withdraw their money at any point in time, there are a couple of approaches to model the savings volume. The liquidity maturity can either be estimated by an outflow model, which only estimates the liquidity profile of savings for current clients, or by a going concern model, where the development of the entire savings portfolio is estimated. The outflow models ignore new clients and inflows from current clients, and therefore do not provide an estimate of the expected future savings volume.

Often, the periodic outflow is estimated with the help of so-called vintage analysis. This is illustrated in Figure 2. With this technique, snapshots at different moments in time are taken from the savings portfolio. Each snapshot contains a group of clients that form a vintage, for which the periodic outflow is measured (see the example figure below). This time effect is caused by a relatively high outflow from unstable clients with volatile savings balances, after which a stable core remains.

Next to the average outflows, certain patterns are sometimes observed in the series of outflows and these can be included as explanatory variables. Examples of these include a decay factor, seasonal effects, the impact of competitors’ savings rates and/or macro-economic variables.

The ’going-concern’ model and the interest rate risk sensitive model include future inflow of volume to a certain extent. Under the ‘going concern’ principle, volume models estimate the development of the entire savings portfolio. Under this model, both savings inflow and outflow from both existing and new clients is included. From an interest rate risk perspective, the difference between the volume and volume outflow model can be used to determine which part is assumed interest rate risk sensitive. This yields a consistent approach over the balance sheet (e.g., corporate loans and mortgages). Future inflows that do not enter the balance sheet at neutral market value (par) are considered for IRR purposes. The disadvantage is that estimates are needed to determine the repricing speed of NMP volume.

 

Figure 2: Illustrative example of a vintage analysis.

Figure 2: Illustrative example of a vintage analysis.

Relevance of segmentation

In modeling volumes and client rates, the savings portfolio segmentation is important. The degree of segmentation depends on the materiality and heterogeneity of the savings. This is because the attitude to interest rate risk and liquidity spread risk varies between different product types and client groups. Even within client groups, certain factors may require further segmentation. This could include the level of savings, for example, due to the coverage limit of the deposit guarantee scheme.

Fit for model and purpose

Behavioral modeling 
The client rate model forecasts the relationship between the market and liquidity spreads and the bank’s client rate. The volume model forecasts the savings volume based on either a run-off, going-concern or interest rate sensitivity-based assumption. Therefore, fit for behavioral modeling purposes.

Value (IRRBB)
Suited for the calculation of non-key rate value metrics (market value, BPV, duration). Less suited for key rate duration calculations due to unexpected outcomes for certain tenors.

Earnings (IRRBB)
Client interest expenses can be forecasted using the client rate model, using scenarios for the market rates and liquidity spreads. Earnings can be forecasted by the difference between forecasted investment income and client interest expenses. Therefore fit for earnings measurement and management purpose.

Stress testing
Shocked market rate scenarios can be translated towards shocked client rate forecasts through the client rate model. Therefore the cash flow model can be used for both earnings- and value-based stress testing.

Liquidity spread risk
Fit for liquidity spread risk measurement and management purpose as liquidity spreads can be included as a driver in the client rate model.

Internal risk transfer
The cash flows resulting from the model do not translate into a portfolio with fixed rate instruments constructed with an optimal margin objective. Therefore it is not fit for internal risk transfer purposes.

Benchmarking figures (around 15 Dutch and Swiss banks, including G-SIBs and D-SIBs)

  • 100% use the model for value and earnings calculations and stress testing purposes
  • 75% use the model for Earnings@Risk purposes
  • 25% use the model for FTP purposes
  • 25% use the model for liquidity spread risk purposes

Replicating portfolio

There are many banks that use a framework of replicating investment portfolios to measure and manage the interest rate risk of variable savings deposits. A replicating investment portfolio is a collection of fixed-income investments that aims to reflect the typical interest rate maturity of the non-maturing savings deposits. The optimal portfolio is typically determined by optimizing the margin, the difference between the investment proceeds and the client rate. Banks typically use a margin level or stability-based objective to design the optimal portfolio. This means that the investment strategy is formulated so that the margin between the portfolio return and the savings interest rate is as stable as possible, given various scenarios. 

There are two commonly used methodologies, known as the marginal investment strategy and the portfolio investment strategy. Marginal investment rules are typically used for stable FTP values, while steady state investment rules are used for immediate convergence to the estimated interest rate risk position from a value perspective.

A replicating framework enables banks to base interest rate risk measurement and management on investments with a fixed maturity and price, while the deposits have no contractual maturity or price. Banks can use the model framework to transfer the interest rate risk from the business lines to the central treasury department, by turning the investments into contractual obligations.

Figure 3: Translation of a historical balance overview to a replication portfolio

Figure 3: Translation of a historical balance overview to a replication portfolio

A split between stable and non-stable volume

Due to the volatile behaviour of on-demand savings volume over the months, volume can be split into a stable and non-stable layer. In the replicating portfolio framework, the sum of all volume above a certain portfolio specific layer threshold, the non-stable volume, is monthly (re)invested in market instruments with a 1-month maturity. The stable volume layer contains the remaining volume and is invested in accordance with a replicating portfolio rule. The replicating portfolio rule is constructed such that the maturity profile accurately translates market rate changes to client rates. The portfolio is rebalanced on a monthly basis by the generation of new instruments, to conform to the maturity profile from the replicating portfolio rule.

Considerations for model estimation

Banks should determine the optimization objective in the replicating portfolio construction process. This is often either to achieve margin stability, a maximum margin level, or a combination of both, under different interest rate scenarios. Replicating portfolio calibration can be backward-looking or forward-looking. With backward-looking calibration, historical client rates and market investment returns are used in the optimization. Using forward-looking calibration, different market rates and client rate scenarios may be considered in the optimization process, based on expert input. A bank can also opt to apply liquidity constraints in the optimization process, via a maximum investment horizon and minimum allocations to short-term tenors in the investment portfolio.

Fit for model and purpose

Behavioral modeling
Not fit for behavioral modeling. Although the replicating portfolio maturity profile translates market rate changes to client rates, the client rate is not explicitly modeled in terms of market rates and liquidity spreads. Similarly, although liquidity constraints can be incorporated in the optimization, volume is not directly modeled in terms of its explanatory variables.

Value (IRRBB)
Fit for value measurement and management. Value measures can be calculated based on the (discounted) cash flows from the replicating portfolio. Because the behavioral components are not explicitly being modeled, the interest rate sensitivity in client behavior is not included in value measurement.

Earnings (IRRBB)
Earnings can be forecasted by the difference between forecasted investment income and client interest expenses. It is difficult to use the model to project client interest expenses as the behavioral components are not explicitly modeled.

Stress testing
Suited for stress testing on value measures derived from the replicating portfolio. It is less suited for stress testing on earnings measures as there is no explicit client rate model.

Liquidity spread risk
Not fit for liquidity spread risk measurement as liquidity spreads are not explicitly modeled in the replicating portfolio.

Internal risk transfer
Fit for internal risk transfer, as the model output with instruments with fixed coupons and fixed maturities can be used directly to transfer the interest rate risk to treasury (FTP).

Benchmarking figures (around 15 Dutch and Swiss banks, including G-SIBs and D-SIBs)

  • 100% use the model for value and earnings calculations and stress testing purposes
  • 43% use the model for Earnings@Risk purposes
  • 86% use the model for FTP purposes
  • 25% use the model for liquidity spread risk purposes

Hybrid model

Some banks use a hybrid model for their savings portfolios. This model uses both the cash flow model and the replicating portfolio for optimal alignment with purpose and use. Typically, the client rate model is used for earnings measurement, and the replicating portfolio for value measurement.

The cash flow model models the behavioural aspects of volume and client rate with respect to interest rate risk. The BPV/duration resulting from the cash flow model is used as constraint for the replicating portfolio construction. The replicating portfolio replicates the value-based interest rate risk measure (BPV/duration) from the cash flow model. Monthly rebalancing of the portfolio is done by generating new instruments to conform to maturity profile from replicating portfolio rule.

Figure overview: Schematic overview of how the model concepts interact in a hybrid model.

Figure overview: Schematic overview of how the model concepts interact in a hybrid model.

Fit for model and purpose

Behavioral modeling
Behavioral modeling is included in the cash flow model, but the value measures are typically based on the replicating portfolio. Therefore the hybrid model does not explicitly model the interest rate sensitivity of the behavioral components in a value context.

Value (IRRBB)
Fit for purpose as value measures can be calculated based on the (discounted) cash flows from the replicating portfolio. The interest rate sensitivity in client behavior is not included in value measurement.

Earnings (IRRBB)
Fit for purpose as client interest expenses can be forecast using the client rate model, using scenarios for the market rates and liquidity spreads. Earnings can be forecast by the difference between forecasted investment income and client interest expenses.

Stress testing
Both the replicating portfolio (for value measurement purposes) and client rate model (for earnings measurement purposes) can calculate the effect of shocked market scenarios. Therefore, the model is suited for both earnings- and value-based stress testing.

Liquidity spread risk
Liquidity spreads can be included as driver in the client rate model (earnings measurement purposes), however they are not explicitly modeled in the replicating portfolio model (value measurement purposes).

Internal risk transfer
Fit for purpose as the model output with instruments with fixed coupons and fixed maturities can directly be used to transfer the interest rate risk to treasury (FTP).

Benchmarking figures (around 15 Dutch and Swiss banks, including G-SIBs and D-SIBs)

  • 100% use the model for value and earnings calculations and stress testing purposes
  • 100% use the model for Earnings@Risk purposes
  • 100% use the model for FTP purposes
  • 75% use the model for liquidity spread risk