BPCE_PILLAR_III_2017

CREDIT RISK Internal ratings and risk measurements

Backtesting All information used to measure the Group’s exposure to all counterparties bearing credit risk is saved. Furthermore, all information on counterparties in default (collections, deterioration, write-offs) for the period is archived. Validity tests are conducted once a year on each of these internally-estimated inputs. All three credit risk inputs are subject to backtesting each year in order to verify the performance of the rating system. More specifically, backtesting is aimed at measuring the overall performance of models used, primarily to ensure that the model’s discriminating power has not declined significantly relative to the modeling period. The average of estimated and observed values is calculatedover several years using the informationavailablefor each model. Observed default rates are then compared with estimated default rates for each rating. Ratings are checked for through-the-cycleapplicability.More specifically, for portfolios with low default rates (large corporates,banks, sovereignsand specialized lending), a detailed analysis is carried out using additional indicators such as severity differences, adjustments to agency ratings and changes in ratings before default. A more qualitative analysis is also performed.

The scope of loss given default values is consistent with the values observed, i.e. limited exclusively to exposures at default. Estimated values therefore cannot be directly compared with LGD values measured in the outstandingportfolio. The average of estimated and observedvalues is calculatedover several years using the information available for each model. Actual collections are compared with estimated collections. Downturn LGDs are also verified. Backtestingresults may call for the implementationof action plans if the system is deemed not sufficiently prudent or effective. Backtesting results and the associated action plans are discussed by the Group Modeling Committee,then reviewed by the DRCCP Group Standards & Methods Committee (see governance of the internal rating system). On the basis of these exercises, the rating system has been deemed satisfactoryon the whole in terms of effective risk management.The calibrations are conservative with respect to observed risk: default rates observed are lower than the default rates expected over the entire cycle and over the most recent period. Losses observed on assets indefault arelower thanexpected losses.

5

87

Risk Report Pillar III 2017