CCFA Risk Study Group Review – Session 4: Enterprise Wide Stress Testing

12 Aug 2015 10:57 PM | Siyao He (Administrator)

After the delivery of Fundamental Review of the Trading Book on April 16, 2015, CCFA Risk Study group held the fourth study session. The topic is Enterprise Wide Stress Testing (EWST) on August 12, 2015. It was an honour to invite three experienced practitioners in the field of stress testing to lead the round table discussion. They are: Xiaobo Wang, Director in Liquidity Management at Scotiabank, Jeff Xue, Senior Director from CIBC and Beizhen Lei, Senior Manager of BMO Model Risk Validation. Yicent Chen, Model Risk Specialist from BMO was the facilitator. Other practitioners from leading Canadian financial institutions also joined the discussion and exchanged their views on the industry practice.

The discussion started from a high-level overview of the stress testing practice, including the objectives, approaches and components. Stress testing is a key risk and business management tool to identify and quantify key risks and assess capital adequacy during stressed periods. There are two approaches commonly used for stress testing: A top down approach starts with a systematic adverse scenario defined by the regulator and assesses the impact of such scenario on bank’s financial statements. This approach enables the regulator to assess impacts across different banks under adverse scenario. On the other hand, a reversed approach starts with defining a severely stressed outcome of the bank and then identifies the scenarios that would cause such severe loss. This enables a bank to tailor the stress testing practice to its unique business exposures. Components of stress testing include risk components (credit risk, counterparty credit risk and market risk) as well as stress scenarios (US recession, Emerging Markets, crude oil, etc.).

Then the discussion leads to a closer look on the methodology of corporate credit stress testing, especially the methodology of Probability of Default (PD) stress testing. Stress testing models are built with internal or external historical rating and default data. The conditional transaction matrix approach is commonly used for PD stress testing. Under this approach, first transition matrices are constructed with chosen horizon based on historical data. Then a summary statistic, such as downgrade and default probability (DDP) is constructed to measure the changes in the transitions. After that an econometric model is developed to relate the summary statistic to relevant macroeconomic/financial variables. Finally a stress transition matrix is derived corresponding to the model predicted DDP for a macroeconomic scenario.

The discussion concludes with some cases studies in actual stress testing practice of the banks. It is pointed out that stress testing should be viewed as a system instead of individual components/models. By the end of the day, stress testing seeks to answer the question of financial statements projection under adverse scenarios. Knowledge of business, accounting, strategic planning as well as risk management should be integrated for a meaningful stress testing practice. Stress scenario could potentially affect both asset and liability side on the balance sheet. Output analysis is valuable in the aspect of assessing out-of-sample model performance as well as soundness of model methodology.

The participants in the discussion also exchanged opinions on data collection, modelling methodology and implementation of stress testing. Topic for next study session will be operational risk.    

在就交易账户(trading book)市场风险研讨后,CCFA2015812日迎来了第四次风险管理研讨会,课题是企业整体压力测试。我们有幸请到了在这一领域的几位资深从业者领导讨论,他们是Scotia Bank流动性风险管理主管Xiaobo WangCIBC高级主管Jeff XueBMO高级经理Beizhen LeiYicent Chen担任本次讨论的主持。来自各大金融机构其他与会者也就本议题参与讨论,提供专业见解。


随后,与会者就信用风险的压力测试,即违约概率的压力测试进行了深入讨论。压力测试模型是基于金融机构内部或者外部的历史数据构建的。条件转移矩阵是业界常用的违约概率压力测试的建模方法:首先在给定历史时期,基于历史数据我们可以得到信用评级的转移矩阵。该矩阵可以被一个代表性的变量,降级和违约概率(downgrade and default probability)所表征。随后我们建立计量模型,使用该变量随后作为被解释变量, 宏观或金融数据作为解释变量,并进行预测。最后通过预测值,我们可以推导出在压力情景下的条件转移矩阵。


与会者也就数据采集,模型方法和实施等方面交换了意见。 下次研讨会的议题是运营风险。

Powered by Wild Apricot Membership Software