NATIXIS // 2021 Universal Registration Document

RISK FACTORS, RISK MANAGEMENT AND PILLAR III Risk management

The indicators defined for backtesting are used both to validate the model and measure its performance. Two main types of indicators are used: population stability indicators: these analyses are used to check V that the population observed is still similar to the population that was used to build the model. The model may be called into question if the segmentation variables or the LGDs result in excessively large distribution differences. All of these indicators are compared with the benchmark indicators (usually those calculated when the model was built or those from external data or agencies). These analyses are applicable to both expert appraisal-based models and statistical models; model performance indicators: the model’s performance is V measured to validate the segmentation and also to quantify, overall, the differences between the forecast and actual figures. This is achieved by using statistical indicators, which are compared against those calculated during modeling. Loss given default models (internal LGD) are calculated:

on expert models based on internal and external histories and V external benchmarks for banks and sovereigns; on a statistical basis for the corporate asset class; V using stochastic models if there is a claim against a financial V asset. The results of the backtesting may require recalibrating the risk input, where appropriate. Once complete, a backtesting report is produced to provide: all the results of the backtesting indicators used; V any additional analyses; V an overall opinion of the results in accordance with the Group’s V standards. The report is then submitted to the internal validation teams (Model Risk Management)for their input, and subsequentlypresented to the various Committees to inform the bank’s management.

3

Backtesting of LGDs and PDs by exposure class

2021 backtesting figures

Observed default rate

Observed LGDs

Model LGDs

Estimated PD

Sovereigns

40.20% 28.41% 29.04%

65.20% 59.78% 32.57%

0.37% 0.11% 0.61%

2.23% 0.52% 0.66%

Banks

Corporate

This table provides a general summary of the system’s performance but differs from the annual backtests carried out within the Group, which are conducted on a model-by-model basis and not overall by portfolio. However, this table allows a comparison of estimates and actual results for each internal input over an extended period and for a significant, representative percentage of each exposure class. The results come from data warehouses used for modeling. This is based on all performing customers for default rates and PD, and on all customers in default for LGD. These results also take into account the latest regulatorychanges (guidanceon probabilityof default (PD) estimates and loss given default (LGD) estimates). These checks are conducted by backtesting the various rating models once a year by scope and the results are presented to the Credit Risk Model Monitoring Committee (CRMMC) which meets at least quarterly. Subsequently, the results are submitted to the internal validation team (Model Risk Management) and presented to the various Committees in order to inform the bank’s Management. The CRMMC Committee: serves as a forum for presenting the results of performance and V stability measurements; analyzes the indicators whose alert thresholds have been V exceeded; decides on any measures to be taken to correct any deviations or V anomalies. These measures may take different forms, including changing rating practices, methodologies, performance analyses or alert threshold values.

Monitoring and backtesting of counterparty risk model Backtesting and validation are key items of governance in the Internal Model Method approach. In accordance with general regulatory requirements, the reliability of internal models must be monitored regularly using a comprehensive backtesting program. This process is essential for ensuring the quality and relevanceof the results obtained from the models that have been developed and used for both internal risk management, and to meet regulatory obligations. The counterparty risk backtesting program is designed to validate the key assumptions on which the exposure model is built – stochastic processes for market risk factors, correlationsand pricing models – and to identify notable discrepancies in specific model elements. The framework developed is based on the following two backtests: market risk factor backtesting, i.e. to assess the predictivecapacity V of the stochastic processes used to describe the dynamics of unique risk factors; portfolio backtesting: i.e. to assess the full exposure model V (stochastic process, correlations and pricing) for portfolios representing Natixis’ exposures. In respect of the market risk factor, backtesting uses historical trends to test risk factor predictions based on the stochastic processes. Backtesting can be performed ex-post by taking the aggregate historical market data for the selected backtestingperiod, or, to expand the results over wider time horizons and/or cover a broader range of market conditions. It may also be performed retroactively: this specific approach is called retro-backtesting. Here, the aim is to demonstrate that the models would have worked correctly had they been implemented in earlier periods and are thus suitable for future use.

127

www.natixis.com

NATIXIS UNIVERSAL REGISTRATION DOCUMENT 2021