PRESS RELEASE – December 1, 2020
Latest report from Global Credit Data highlights need for unresolved defaults to be incorporated into modelling process
As the COVID-19 pandemic crisis continues, banks still face an uncertain impact on their credit risk assessment. Now, more than ever, it is of utmost importance to incorporate recent information into credit risk models and especially Loss Given Default (LGD) models. If these models are based on historical defaults with complete workout profiles only, most recent developments are naturally not included.This creates a resolution bias which leads to underestimating recent losses, since only quickly resolved, cured cases (typically with lower losses) can be included. Hence, average LGDs based only on resolved loans can result in unrealistic long-term average values. Knowing this, regulatory authorities explicitly ask for unresolved defaults to be incorporated into the modelling process (e.g. European EBA GL 2017/16 or US BCC Bulletin 13-5).
Establishing how to account for incomplete recovery processes or unresolved loans is therefore a key question for credit risk modellers right now. This report describes the GCD methodology for calculating LGDs for unresolved loans. The methodology benefits from GCD’s detailed and granular collection of post-default cash flow data and is based on extrapolations of historical recovery cash flows refined by the usage of risk drivers.
GCD’s data is detailed enough to develop and enhance internal LGD models and to be used for validation, calibration or benchmarking purposes. These models can be used to support the Advanced Internal Ratings-Based (AIRB) approach, to fulfil the credit provisioning standards IFRS9 or CECL, as well as for stress-testing, economic capital and pricing, among other uses.
The methodology has been developed on the entire GCD database. For the sake of simplicity, this report presents the results for Large Corporates only (defined by their sales or assets being above €50m). Results for the rest of the segments are available upon request.
The key observations are:
• Including unresolved default cases when assessing LGD leads to a more conservative long-run average LGD, although this effect is overall limited to 2%.
• The impact becomes more significant when looking at the most recent years – including unresolved defaults causes an increase in the long-run average LGD by 12% for 2018.
• For default years before 2013 or more than seven years in default, the completion rate of defaults is above 95% and these years can therefore be considered complete.
GCD provides unresolved LGDs to its member banks both at a loan and borrower level. This allows member banks to calculate long-run average LGDs for calibration purposes, avoiding a resolution bias when using most recent default cases.
“The methodology provides a straightforward, data-driven way of incorporating incomplete workout processes in the estimation of longrun average LGDs. Extensive validation both in- and out-of-sample has shown that the method works well in predicting LGDs for unresolved defaults,” says Nunzia Rainone, Executive at GCD.