In this presentation I gave in the 7th Annual Risk Americas Conference, I first discussed the inconsistency of CECL from risk philosophy perspective, and then shared some thoughts on key aspects of CECL modeling, i.e. Reasonable and Supportable Period, leveraging CCAR models for CECL, and model performance testing.
1. Current Expected Credit Loss (CECL)
Modeling
May 17th 2018
Xiaoling (Sean) Yu
SVP, Director of Model Validation
KeyBank
7th Annual Risk Americas 2018
2. 2
Disclaimer
7th Annual Risk Americas 2018
The views expressed in this presentation are those of the
speaker and do not necessarily reflect the views of
KeyCorp or its subsidiaries in all respects.
3. 3
What is CECL?
7th Annual Risk Americas 2018
⢠On June 16, 2016, the FASB issued ASU No. 2016-13, Financial InstrumentsâCredit
Losses (Topic 326). The new guidance requires organizations to measure all expected
credit losses for financial instruments held at the reporting date based on historical
experience, current conditions and reasonable and supportable forecasts.
o To eliminate the probable initial recognition threshold and the concept of âLoss Emergence Periodâ
- all expected credit losses over the contractual term
o To change from a backward-looking to a forward-looking approach
o Beyond the period for which reasonable and supportable (R&S) forecasts are obtainable, banks
may rely on historical information alone
⢠CECL is aimed to enable more timely recognition of credit losses, as the existing approach,
based on a âprobableâ threshold and an âincurredâ notion, resulted in loan loss allowances
that were âtoo little, too lateâ in the period leading up to the 2008 financial crisis.
4. 4
Risk Philosophy of CECL
7th Annual Risk Americas 2018
Future losses are uncertain and generally defined along three dimensions:
⢠Time Horizon (e.g. 12 months, 9 quarters, etc.)
⢠Distribution â Expected (EL) vs Unexpected (UL)
⢠Cycle Sensitivity â Conditional (PIT) vs Unconditional (TTC)
PIT TTC
EL
UL
Econ.
Path
R&S
CECL is life time expected
loss that reflects a hybrid view
on cycle sensitivity.
Y1 Y2 Y3 Y4 Y5 Y6 Y7
Reversion Historical Avg.
5. 5
Leveraging Existing Credit Risk Models?
7th Annual Risk Americas 2018
CECL Modeling Needs
Existing Models
Incurred ALLL Basel AIRB Economic Capital CCAR
PD
⢠Conditional
⢠Expected Lifetime
⢠Conditional but
backward looking
⢠Loss Emergence
Period
⢠Unconditional
⢠12 Months
⢠Unconditional but with
correlation assumptions
⢠12 Months for Default Risk,
but Expected Lifetime for
Full Credit Migration Risk
⢠Conditional
⢠9 Quarters
⢠Conservative
LGD ⢠Conditional ⢠Unconditional ⢠Downturn LGD
⢠Unconditional but may with
correlation assumption
⢠Conditional
⢠Conservative
EAD
⢠Existing Portfolio
⢠Conditional LEQ
⢠Conditional
Prepayment
⢠Existing Portfolio
⢠Unconditional LEQ
⢠No Consideration of
Prepayment
⢠Existing Portfolio
⢠Unconditional LEQ
⢠No Consideration of
Prepayment
⢠Existing Portfolio
⢠Unconditional LEQ
⢠Unconditional Prepayment
⢠Existing Portfolio &
New Volume
⢠Conditional LEQ
⢠Prepayment is
typically ignored
Macro-
Econ.
Forecast
⢠Reasonable and
Supportable
⢠Reverting to
Historical Average
N/A N/A
⢠Indirectly Assumed
/Simulated
⢠Several Selected
Scenarios
6. 6
Reasonable and Supportable (R&S) Period?
7th Annual Risk Americas 2018
Para 326-20-30-9:
⢠âWhen an entity uses historical loss information, it shall consider the need to adjust historical
information to reflect the extent to which management expects current conditions and reasonable and
supportable forecasts to differ from the conditions that existed for the period over which historical
information was evaluated.â
⢠âAn entity shall not adjust historical loss information for existing economic conditions or expectations of
future economic conditions for periods that are beyond the reasonable and supportable period.â
Quantitative Interpretation:
⢠The R&S Period may be defined by the acceptable level of model error. For instance, R&S period can
be determined when further extending it will not lead to material change in the level of forecast error.
⢠R&S Period could be product/asset specific, and/or cycle specific.
⢠The model error consists of two components: the error in macroeconomic forecast and the error in loss
forecast conditional on given macroeconomic variables.
⢠Economists do not have a good track record in forecasting recessions, and the range of forecast
variances can be large.
7. 7
Reasonable and Supportable (R&S) Period? (contâd)
7th Annual Risk Americas 2018
⢠The Federal Open Market
Committee (FOMC)âs Summary of
Economic Projections (SEP) reports
the root mean squared errors
(RMSEs) of real-time forecasts over
the past 20 years made by a group of
leading private and public sector
forecasters.
⢠Shaded bands show median SEP
forecasts (as of September 2016) Âą
average historical RMSE at the
appropriate forecast horizon, which
cover approximately 70% future
outcomes assuming future prediction
errors are stable, unbiased and
normally distributed.Source: Reifschneider and Tulip (2017)
8. 8
Reasonable and Supportable (R&S) Period? (contâd)
7th Annual Risk Americas 2018
Qualitative Interpretation:
⢠R&S period is more of a process and governance concept instead of a quantitative concept.
⢠It has to be consistent with the time horizons of other internal processes, e.g. strategic planning.
⢠R&S period has to be long enough to capture the near future loss, especially when entering into a
recession.
Current Practices:
⢠Majority of the institutions are considering 2-3 years of R&S period, while others may select either one
year or 5-6 years.
⢠Majority of the institutions are considering to apply the same R&S period across asset classes.
Conclusion:
⢠R&S period cannot be a pure quantitative or qualitative concept.
⢠R&S period has to be consistent with the time horizons of other processes across the organization.
⢠The rationale and process followed in determining R&S period should be clearly documented, with
transparent discussion on its quantitative implication.
⢠R&S period has to be periodically reviewed and challenged.
9. 9
Modeling Approach
7th Annual Risk Americas 2018
⢠CECL does not require specific approaches when developing the estimate of expected
credit losses.
⢠Commonly used CCAR modeling approaches, e.g. Hazard Rate model for retail and
Transition Matrix Model for commercial, can be leveraged in principle.
⢠However, CCAR models tend to produce conservative outcomes, especially under base
scenario. How to design proper adjustment has to be thought through.
⢠In addition, efforts need to be made to ensure CCAR models are adjusted to meet CECL
requirements.
o ASU requires the estimate to be based on a financial assetâs amortized cost. If only the unpaid
principal balance (UPB) write-offs are considered in loss history, adjustments would need to be
considered for LGD and EAD modeling.
o While prepayments, extensions, renewals, and modifications are typically not well considered in
commercial CCAR modeling, they need to be carefully examined for CECL.
10. 10
Macroeconomic Variable Selection
7th Annual Risk Americas 2018
CCAR:
⢠Fed scenarios specifies 16 domestic macroeconomic variables that measure economic activity and
prices, asset prices or financial conditions, and interest rates.
⢠CCAR models typically leverage many more macroeconomic variables to capture portfolio specific
sensitivities to different regions, markets, asset types, etc.
⢠Macroeconomic models are used to translate and expand Fed scenarios. For instance, Moodyâs
Analytics produces a forecast for 1,800 variables based on Fed scenarios.
CECL:
⢠All macroeconomic variables within R&S period have to be forecasted.
⢠Macroeconomic variables should be selected not only by their explanatory power of credit loss, but also
by how well they can be forecasted.
⢠Many specific/granular variables that were extrapolated from fundamental ones may not be fit-for-use.
Conclusion:
⢠CECL models may leverage different macroeconomic variables from those used in CCAR models.
11. 11
Model Performance Testing
7th Annual Risk Americas 2018
⢠Similar as for CCAR models, the accuracy of loss forecast models within the R&S period
can be assessed through back-testing against predefined acceptance criteria.
⢠And the accuracy of macroeconomic forecast model within the R&S period can also be
assessed through back-testing.
⢠However, given that CECL represent a hybrid view on cycle sensitivity, accuracy may not
be an appropriate measure of model performance beyond R&S period.
⢠In addition, since CECL required to significantly expand disclosure about credit risk by
credit quality indicators and years of the assetâs origination (i.e. vintage), the model
performance will need to be assessed at granular level as well.