Société Générale / Risk report 2021 - Pillar 3 2020

6 CREDIT RISK RISK MEASUREMENT AND INTERNAL RATINGS PILLAR 3 - 2021 | SOCIETE GENERALE GROUP | 88 TABLE 33: MAIN CHARACTERISTICS OF MODELS ANDMETHODS USED – RETAIL CLIENTS Parameter modeled Portfolio/Category of Basel assets Number of models Methodology Number of years of default/loss RETAIL CLIENTS Residential real estate 8 models according to entity, type of guarantee (security, mortgage), type of counterparty: individuals or professionals/VSB, real estate investment company (SCI). Statistical model (regression), behavioural score. Defaults observed over a period of more than 5 years. Probability of Default (PD) Other loans to individual customers 15 models according to entity and to the nature and object of the loan: personal loan, consumer loan, car loan, etc. Statistical model (regression), behavioural score. Defaults observed over a period of more than 5 years. Renewable exposures 5 models according to entity and nature of the loan: overdraft on current account, revolving credit or consumer loan. Statistical model (regression), behavioural score. Defaults observed over a period of more than 5 years. Professionals and very small businesses (VSB) 10 models according to entity, nature of the loan (medium- and long-term investment credits, short-term credit, car loans), and type of counterparty (individual or real estate investment company (SCI)). Statistical model (regression or segmentation), behavioural score. Defaults observed over a period of more than 5 years. Residential real estate 8 models according to entity, type of guarantee (security, mortgage), and type of counterparty: individuals or professionals/VSB, real estate investment company (SCI). Statistical model of expected recoverable flows based on the current flows. Model adjusted by expert opinions if necessary. Losses and recoverable flows observed over a period of more than 10 years. Loss Given Default (LGD) Other loans to individual customers 17 models according to entity and to the nature and object of the loan: personal loan, consumer loan, car loan, etc. Statistical model of expected recoverable flows based on the current flows. Model adjusted by expert opinions if necessary. Losses and recoverable flows observed over a period of more than 10 years. Renewable exposures 7 models according to entity and nature of the loan: overdraft on current account, revolving credit or consumer loan. Statistical model of expected recoverable flows based on the current flows. Model adjusted by expert opinions if necessary. Losses and recoverable flows observed over a period of more than 10 years. Professionals and very small businesses 12 models according to entity, nature of the loan (medium- and long-term investment credits, short-term credit, car loans), and type of counterparty (individual or real- estate investment company (SCI)). Statistical model of expected recoverable flows based on the current flows. Model adjusted by expert opinions if necessary. Losses and recoverable flows observed over a period of more than 10 years. Credit Conversion Factor (CCF) Renewable exposures 12 calibrations by entity for revolving products and personal overdrafts. Models calibrated by segment over a period of observation of defaults of more than 5 years. MONITORING THE PERFORMANCE OF INTERNAL MODELS The performance level of the entire retail client credit system is measured by regular backtests, which check the performance of PD, LGD and CCF models and compare estimated figures with actual figures. Each year, the average long-term default rate observed for each homogeneous risk pool is compared with the PD. If necessary, the calibrations of PD are adjusted to preserve a satisfactory safety margin. The discrimination level of the models and changes in the portfolio’s composition are also measured. annual default rate). The historical default rate has been calculated on the basis of performing exposure over the period from 2010 to 2019. Credit customers are included in accordance with the revised instructions of the EBA publication of 14 December 2016 (EBA/GL/2016/11). The results presented above cover the entire Group portfolios. Backtests compare the estimated probability of default (arithmetic mean weighted by debtors) with the observed results (the historical While the historical default rate is relatively stable across all Basel Retail portfolios, the probability of default is falling. Indeed, the quality of the counterparties has artificially improved (migration of populations to the best risk classes) due in particular to government measures taken in the Covid context. As the risk scores are mainly based on the number of days of non-payment and the cash balance, they are impacted by: (i) the freezing of the number of days of non-payment and (ii) the artificial increase in cash balances by the payment of government guaranteed loans.