An overview on the proposed estimation methods. Bernhard Eder / Obergurgl. Department of Banking and Finance University of Innsbruck

Transcription

1 An overview on the proposed estimation methods Department of Banking and Finance University of Innsbruck / Obergurgl

2 Outline

3 Impairment of financial instruments Financial instruments measured at AC or FV-OCI Previously: IAS 39 - Backward-looking impairment model Objective evidence Adverse credit events Now: IFRS 9 - Forward-looking impairment model Recognize 12-month expected credit loss, when there is no significant increase in credit risk Recognize lifetime expected credit loss, otherwise "Significant" statistically significant

4 Impairment of financial instruments Expected

5 Impairment of financial instruments Expected Principle-based standard How to operationalize significant increases in credit risk? How to estimate (lifetime) expected credit losses?

6 Assessing the current state of knowledge

7 Assessing the current state of knowledge

8 Assessing the current state of knowledge Accountants are unfamiliar with the statistical methods to estimate expected credit losses.

9 Assessing the current state of knowledge

10 Assessing the current state of knowledge Sources of non-compliance and obscurities Idiosyncratic approaches (loosely motivated, few references) Missing information on data,, robustness checks and validation

11 Expected loss parameter (CRP) approach [ T EL i = E t=1 EAD i (t) LGD i (t) PD i (t) ] 1 (1 + r i ) t X 0 Not mandatory to use credit risk parameter approach Difference EL and ECL: interest rate adjustment

12 PD term-structure Portfolio Level Class Subclass Type Appl. Ref. Moving Average 1 3 Charge-Off Portfolio Time Series 1 1 Survival Vintage - 4 Linear model (discr. time) - - Logistic regression (discr. time) 7 14 APC- 3 4 Retail Non-Parametric Kaplan-Meier, Nelson-Aalen, Breslow - 2 Survival Semi-Parameric Proportional hazard - 3 Account Proportional hazard - 5 Parametric Accelerated Failure Time - - Mixture Cure Rate - 2 Competing Risk Models - 2 Frailty Models 1 1 Matrix Discrete Roll Rate 2 7 Continous Intensity Models - - Homogenous Discrete (cohort) Inhomogenous - 2 Matrix (reduced form) Homogenous 2 7 Continous (intensity) Wholesale Customer Inhomogenous 1 3 Market (structural form) Merton 1 9 CDS 1 8 Agency-Replication 1 1

13 Estimation of expected credit losses Sources of variability There is no single-best method to estimate ECL Fair comparisons require Out-of-time validation Distinction between loan- and portfolio level Substantial scope of discretion Stage transfer Model type, variable selection, and calibration Portfolio segmentation Internal, historical data at banks is limited Provisions will be difficult to compare across banks

14 Other results Bucketing process Metric, scaling (additive, relative, combined) and threshold Confusion about interpretation Downturn adjustments of LGD estimates Margin of conservatism Reliable estimation of parameter correlations Technical consideration of "forward-looking information" Ambiguous wording, e.g. bias and significance

15 Literautre Modest attention in academia Gap in the accounting literature Quality could be improved Motivation, explanation, references to prior literature Data description, robustness checks, (out-of-time) validation Substantial scope of discretion Many are available but availability of historical data is limited Assessment of IFRS 9 Stage transfer makes ECLM (unduly) complex Interpretation difficulties due to principle-based language