Five modelling challenges in modern forecasting and stress testing

Since the implementation of International Financial Reporting Standard (IFRS) 9, capital & impairment forecasting and stress testing has seen a sharp rise in complexity across the banking industry. 4most consultants have been working on projects dealing with Internal Capital Adequacy Assessment Process (ICAAP), regulatory systemic stress tests and planning forecasts. Here are the top five challenges we’ve faced with clients:

1.       Nested time series
Under IFRS 9, impairment modelling involves time series curves that span the lifetime of each asset, with potential ‘sub-series’ for prepayment, time to default / cure and other model components. Forecasting this impairment calculation wraps another layer of time series around the existing levels. Compared to the previous generation of forecasting models, with simple arrears segmented balance forecasts and associated coverage rates, this is a substantial step up in conceptual complexity. Forecasting teams will need to factor into their planning the likelihood of difficulties converting designs into reality and the expertise needed to explain the new methodology to non-technical stakeholders.

 

2.       Economic scenarios
While forecasts and stress tests previously only needed economic scenarios covering the forecasted months, each of these forecasted months are now the starting point for an additional scenario to enable the impairment calculation, thanks to the nested time series. If the forecast retains the impairment model’s use of multiple probability weighted scenarios this necessitates consideration of discontinuities as they branch away from each other through the forecast. A range of solutions are available, from retaining all this functionality to the simplest approach of assuming perfect economic foresight (this is prescribed for regulatory stress tests) and single scenario IFRS 9 runs.

 

3.       Book quality forecasting
Since actual impairment now relies on much more granular Probability of Default (PD) estimations, this should also be considered in impairment forecasts. The typical approach to achieving this granularity is to forecast the risk grade distribution, which can also be leveraged for capital calculation. However, migration between risk grades is the combined result of all possible grade-to-grade movements, which are too numerous to model directly.

There are established generic methods to overcome this challenge, however these come with weaknesses (e.g. z-shifts, which become excessively tail-heavy beyond a small operating window). Bespoke algorithms can be better tailored to the portfolio being modelled, allowing for more realistic grade distributions through the forecast / stress test.

4.       Stage transfer criteria

Key to the IFRS 9 impairment calculation is whether an exposure has crossed a specified threshold of deterioration in credit quality since origination, which determines whether to provide for 12 month or lifetime expected losses. While this requires account level estimation for actual impairment, this level of granularity is unlikely to be worth the complexity in a forecast. Simpler methods which are still effective at a portfolio level include segmentation by origination PD and using grade migration as the basis for stage migration.

5.       Computational constraints
An IFRS 9 compatible forecast or stress test can be much more computationally intensive than pre-IFRS 9 approaches. For example, a five-year stress test with monthly outputs would be 60 impairment model runs plus the book set-up for each. The computational load can be reduced by a range of simplifications, such as approximating stage migration to run the model at segment level or reducing the forecast frequency below monthly and interpolating.

Over the course of many forecasting and stress testing engagements, 4most consultants have gained a substantial amount of experience building bespoke solutions which address these challenges and link forecasting / stress testing to IFRS 9 to ensure there is one consistent version of the truth.

For more information please visit the Stress Testing page on the right-hand side, or contact Josh Isaac at joshua.isaac@4-most.co.uk.