The forward-looking provisions of IFRS 9

Impact

IFRS 9 will, amongst others, rock the balance sheet, affect business models, risk awareness, processes, analytics, data and systems across several dimensions.

We will name a few related to the financials:

• Transition from IAS 39 to IFRS 9 will lead to a change in the level of provision for credit losses. The transition of the current provisions, which are based only on actual losses and incurred but not reported (IBNR) losses, to an expected loss is likely to have significant impact on shareholder equity, net income and capital ratios.
• P&L volatility is expected to increase after transition, since deterioration in credit quality or changes in expected credit loss will have a direct impact on P&L. The P&L volatility will, however, significantly differ per type of credit portfolio, also depending on counterparty ratings and remaining maturity. Portfolios with loans rated below investment grade will move faster from ‘state 1’ to ‘state 2’ (see box), since a move within investment grade ratings is not seen as a credit quality deterioration. Portfolios with long maturities will face large P&L volatility when moving from state 1 to state 2.
• Capital levels and deal pricing will be affected by the expected provisions.

Total P&L over time will not change, since the expected credit loss provision is booked against the actual credit losses during lifetime. If there is no actual credit loss, all provisions will fall free as profit towards maturity.

Forward-looking

IFRS 9 requires financial institutions to adjust the current backward-looking incurred loss based credit provision into a forward-looking expected credit loss. This sounds logical for an accounting provision and it assumes that existing relevant models within risk management may be applied. However, there are some difficulties to overcome.

Incorporating forward-looking information means moving away from the through-the-cycle approach towards an estimation of the ‘business cycle’ of potential credit losses. A forward-looking expected credit loss calculation should be based on an accurate estimation of current and future probability of default (PD), exposure at default (EAD), loss given default (LGD), and discount factors. Discount factors according to IFRS 9 are based on the effective interest rate; this subject will not be further addressed here. The EAD can mainly be derived from current exposure, contractual cash flows and an estimate of unscheduled repayments and an expectation of the use of undrawn credit limits. Both unscheduled repayments and undrawn amounts are known to be business cycle dependent. Forecasting these items can be derived from historical observations.

Of course, the best calibration is on defaulted data since we determine exposure at default. If insufficient data is available, cycle dependent unscheduled repayments and drawing of credit limits can be derived from the entire credit portfolio, preferably corrected with some expert judgement to reflect the situation at default.

Banks have internal rating models in place to assign a PD to a counterparty and for trenching the portfolio in different levels with a specific PD. From a capital point of view, these ratings are mostly calibrated to a through-the-cycle level of observed defaults. Now using all the bank’s forward-looking information may improve estimates if business cycle(s) can be identified, potential scenarios of the development of the cycle in the future can be forecasted, including how the cycle affects a bank’s PD term structure. This would be a macroeconomic and econometric heaven if there were sufficient data available to derive accurate and statistically significant models. Otherwise, banks need to rely more on expert judgement and external macroeconomic reports.

Next to the PD term structures, LGD term structures are required to calculate a life time expected loss. Deriving an accurate LGD term structure from realized defaults requires a large default database. Deriving a business-cycle dependent LGD term structure requires an even bigger database of accurately and timely documented losses. The level of business cycle dependency of LGD significantly differs per type of counterparty, industry, and collateral. Subordination is not much cycle dependent, while loans covered with collateral, such as mortgage loans, may result in large movements in LGDs over time. Hence, this requires different LGD term structures for different LGD types and levels.

Economic scenarios

Incorporating forward-looking information means modeling business cycle dependency in your PD and LGD. For significant drivers, future scenarios are required to calculate expected credit loss. At most banks, these forward-looking scenarios are commonly the domain of economic research departments. Macroeconomic forecasting concentrates mainly on country-specific variables. Growth of domestic product, unemployment rates, inflation indices and interest rates are typical projected variables.

Usually, only large international banks with an economic research department are able to project consistent economic outlooks and scenarios. Next to macro scenarios, industry specific forecasts are important. Industry risk models enable a bank to make forecasts for a certain industry segment, e.g. chemicals, automotive or oil & gas. Industry models are often based on variables such as market conditions, barriers to entry and default data. At some banks, industries are analyzed and scored by economic researchers. At others, usually smaller banks, industries are ranked by sector business specialists.

Industry scorings often form input for rating models and are important factors for portfolio management purposes. Therefore, caution is required in correlation between drivers of ratings and drivers of the PD term structure.

Credit portfolios

For homogenous retail exposures, forward-looking elements can be considered on a portfolio level by modeling the dependencies of PD and LGD percentages for realized defaults and losses; in essence this is a bottom-up approach. For mortgage portfolios, cycle dependency relates, for example, to unemployment and house price indices, among other factors. However, statistically significant parameters and models for default relations are difficult to obtain since there is a common time gap in observing and administrating both defaults and business cycle.

Model significance can be improved by adding additional variables with increasing risk of overfitting. Even if there is statistical proof for macroeconomic dependencies in PD and LGD rates, it is advised to be cautious, since it also requires designing credible macroeconomic scenarios. As business cycles are difficult to predict, this could lead to extra P&L volatility and an increase in the complexity and ‘explainability’ of figures. Therefore, regular back-testing and continuous monitoring are important for an accurate and robust provision mechanism, especially in the first years after the model is introduced.

For non-retail exposures, country and industry risk are, if embedded in the credit rating models, already part of the annual individual credit review and rating assignment processes. In the monthly financial reporting, additional country and industry risk factors can be taken into account on a portfolio basis, making provisions more forward looking; in essence a top-down approach. If necessary, risk management can make adjustments on an individual basis for wholesale counterparties, and facilities. A forward-looking overlay should improve the accuracy of provisions and a timely and adequate recognition of credit risk, instead of “too little, too late” as under the existing rules.

Governance

Because of the forward-looking character of IFRS 9, and the increasing role of risk models, a transparent and robust governance framework will become more important. Coordination and communication are required across risk, finance, business units, audit and IT.

Risk management typically delivers the expected credit loss parameters and calculations to finance on a monthly basis. Proposals for retail and nonretail adjustments briefly described above, must be discussed and agreed upon, after which the final proposal is submitted to the approval authority.

The governance framework should be documented and reviewed on an annual basis, and highlight key functions, stakeholders, definitions, data management, model (re)development, model implementation, portfolio monitoring and validation. In addition, all parties involved should speak the same credit risk language, have access to detailed data underlying the calculation of the provision and a good under- standing of the model and implications of decisions and parametrization. Only then can the finance department obtain an accurate understanding of the level and change of the provision and clearly inform the board and other stakeholders.

Zanders recommends preparing early for IFRS 9 and having a deep and thorough understanding of the impact, as well as the robust tooling and processes in place. Don’t just wait and ‘watch the hare running’, but start early, and at least run a shadow period during daylight to allow sufficient time.

Hassle-free CECL and IFRS9 compliance? Try our new Condor ECL tool!