X hits on this document





8 / 13


Exhibit 2

Mark-to-model vs. default mode


Mark-to-model vs. default mode for portfolio loss evaluation

Simultaneous evaluation of corporate and retail risk

Separate evaluation of risk

Corporate Retail

5 banks run portfolio on MTM

2 banks run Corporate portfolio on MTM

Mark-to- market/ Mark-to- model

Default- only mode

3 banks run portfolio on DM

2 banks run retail portfolio on DM

  • MtM valuation for corporate portfolio seems the preferred choice

    • “It captures the risk of low-default portfolio accurately”

    • “Measuring the impact of transitions on a long- tenure portfolio is critical for the bank.”

  • The retail portfolio is preferably run in DM, except in banks where it is run together with the corporate portfolio and in MtM mode

Same set of banks with corporate on MtM and retail on DM:

  • Ideal choice of methodology for each component of


  • Diversification benefit between both not recognized

Key modules in a proprietary model include the following:

  • Correlation module. While it makes sense to seek a commercial application, such models do not exist for developing countries. Historical data across multiple business cycles are needed to calibrate a stable correlation model.

  • Complex instrument valuation module. No consensus exists on the most accurate way to model complex instruments in portfolio models.

  • Efficient simulation module. Development of an efficient simulation engine is particularly important for retail portfolios because of the large number of accounts. The banks we interviewed expressed dissatisfaction with the simulation engine underlying KMV products.

Developing and institutionalizing a proprietary model is a multi-year task – typically, it takes at least 1 year and often 2 or more. Several banks have tried to build a full model in one fell swoop, but failed. The lesson here is that it might be better to approach the task in stages, first laying out the specifications and building a prototype model in Excel, and then testing it with real-life data, in order to identify any bugs in the specifications, design and implementation. Rolling out a proprietary model will require buy-in across the organization, not only within the risk team but also from pricing and business development. Testing a prototype model against intuition in a staged approach can help build that support.

Document info
Document views64
Page views65
Page last viewedFri Jan 20 07:47:43 UTC 2017