Société Générale / Risk report 2021 - Pillar 3 2020

9 MARKET RISK MARKET RISK MAIN MEASURES | SOCIETE GENERALE GROUP | PILLAR 3 - 2021 179 The SVaR increased on average in 2020 (EUR 50 million vs EUR 38 million in 2019). Its evolution over the year was marked by three main stages: over the first two months of the year, the SVaR remained at a low p level, following on from 2019; in March, SVaR increased significantly, with a high level of variability p due to the financial crisis triggered by the Covid-19 pandemic. This increase is mainly attributable to the equity scope initially, then mainly results from Fixed Income activities thereafter; the SVaR then fell gradually to reach again in July pre-crisis values p and dynamic. TABLE 92: REGULATORY TEN-DAY 99% SVAR AND ONE-DAY 99% SVAR (In EURm) 31.12.2020 31.12.2019 Stressed VaR (10 days, 99%) (1) Stressed VaR (1 day, 99%) (1) Stressed VaR (10 days, 99%) (1) Stressed VaR (1 day, 99%) (1) Period start 105 33 108 34 Maximum value 343 109 213 67 Average value 158 50 119 38 Minimum value 73 23 49 15 Period end 131 41 112 35 Over the scope for which capital requirements are assessed by internal model. (1) IRC AND CRM At end-2011, Societe Generale received approval from the Regulator to expand its internal market risk modelling system by including IRC (Incremental Risk Charge) and CRM (Comprehensive Risk Measure), for the same scope as for VaR. They estimate the capital charge on debt instruments that is related to rating migration and issuer default risks. These capital charges are incremental, meaning they are added to the charges calculated based on VaR and SVaR. In terms of scope, in compliance with regulatory requirements: IRC is applied to debt instruments, other than securitisations and p the credit correlation portfolio. In particular, this includes bonds, CDS and related derivatives; CRM exclusively covers the correlation portfolio, i.e. CDO tranches p and First-to-Default products (FtD), as well as their hedging using CDS and indices. Societe Generale estimates these capital charges using internal models (1) . These models determine the loss that would be incurred following especially adverse scenarios in terms of rating changes or issuer defaults for the year that follows the calculation date, without ageing the positions. IRC and CRM are calculated with a confidence interval of 99.9%: they represent the highest risk of loss obtained after eliminating 0.1% of the most unfavourable scenarios simulated. category. In addition, a correlation between these five systemic factors is integrated to the model. These correlations, along with the rating transition probabilities, are calibrated from historical data observed over the course of a full economic cycle. In case of change in an issuer’s rating, the decline or improvement in its financial health is modeled by a shock in its credit spread: negative if the rating improves and positive in the opposite case. The price variation associated with each IRC scenario is determined after revaluation of positions via a sensitivity approach, using the delta, the gamma as well as the level of loss in the event of default (Jump to Default), calculated with the market recovery rate for each position. The internal IRC model simulates rating transitions (including default) for each issuer in the portfolio, over a one-year horizon (2) . Issuers are classified into five categories: US-based companies, European companies, companies from other regions, financial institutions and sovereigns. The behaviours of the issuers in each category are correlated with each other through a systemic factor specific to each The CRM model simulates issuer’s rating transitions in the same way as the internal IRC model. In addition, the dissemination of the following risk factors is taken into account by the model: credit spreads; p basis correlations; p recovery rate excluding default (uncertainty about the value of this p rate if the issuer has not defaulted); recovery rate in the event of default (uncertainty about the value of p this rate in case of issuer default); First-to-Default valuation correlation (correlation of the times of p default used for the valuation of the First-to-Default basket). These dissemination models are calibrated from historical data, over a maximum period of ten years. The price variation associated with each CRM scenario is determined thanks to a full repricing of the positions. In addition, the capital charge computed with the CRM model cannot be less than a minimum of 8% of the capital charge determined with the standard method for securitisation positions. The same internal model is used for all portfolios for which an IRC calculation is required. The same is true for the portfolios on which a CRM calculation is performed. Note (1) that the scope covered with internal models (IRC and CRM) is included in the VaR scope: only entities authorised for a VaR calculation via an internal model can use an internal model for IRC and CRM calculation. The use of a constant one-year liquidity horizon means that shocks that are applied to the positions to calculate IRC and CRM, are instantaneous one-year shocks. This (2) hypothesis appears to be the most prudent choice in terms of models and capital, rather than shorter liquidity horizons.