INTERNATIONAL BANKS TO STRENGTHEN USE OF PORTFOLIO CREDIT RISK MODELING SYSTEMS

IIF-ISDA Study Finds Banks View Modeling as Vital to Managing Credit Risk

LONDON, Thursday, February 3, 2000 --- A major international technical assessment of risk modeling systems used by leading banks, was jointly prepared recently by the Institute of International Finance (IIF) and the International Swaps and Derivatives Association (ISDA). Its findings are based on studies of multiple modeling systems used by 25 commercial banks from 10 countries for a range of different types of credit risk exposure.

Leading international banks believe that the use of sophisticated portfolio credit risk measurement models will increase significantly, according to the study. The banks suggest that the efficient and effective use of these systems is of prime importance for the measurement and management of credit risk exposures and to support swift adjustments in bank exposures in response to changing market conditions.

The report seeks to provide a neutral assessment of credit risk modeling so that consideration by the Basle Committee as well as subsequent policy debates can occur on the basis of reliable information. The banks that participated in the IIF-ISDA project believe that their credit risk modeling systems are robustly constructed and, when applied properly, can be extremely useful in managing credit risk. They believe the analysis presented in this report of the study will be helpful to the banks themselves and to banking supervisors in considering how best to implement and evaluate the use of credit risk models.

The study is based both on surveys of banks with regard to qualitative aspects of modeling systems, as well as on a detailed quantitative testing of selected models. On the qualitative front, the report notes that banks' use of modeling systems is likely to increase substantially in the near future. It reports: "When asked whether they intended to implement a new credit risk measurement system in the next 12 months, the majority of banks participating in the corporate bonds (11 out of 19) and middle market (11 out of 17) quantitative exercises indicated that they would be doing so. A significant number of mortgage and small retail credit participants also indicated that they would be implementing new credit risk measurement systems in the next 12 months."

On the quantitative front, the report found that when assumptions, parameters, and portfolios are standardised, outputs are broadly similar within model types (e.g., within Credit Risk+ or within KMV) when the same version of the model was used in the corporate bonds and loans portfolio. In some model types, the outputs are almost identical. The report also notes, however, that the standardised parameters used in this exercise generally are not consistent with any one bank's preferred settings. "When standardised assumptions are relaxed and banks use their own assumptions in their proprietary models, significant differences in outputs should be expected to, and do, result," states the report.

The study underscores the significance to banks in maximizing shareholder value of using advanced credit risk measurement tools such as models. Representatives of the participating banks agreed that implementing sophisticated models demands highly skilled personnel and the investment of substantial resources both to secure excellent data management and to enhance test and recalibrate systems. Important conclusions of the study are as follows:

• Models yield directionally consistent outputs when given similar inputs.
• When assumptions, parameters, and portfolios are standardised, outputs are broadly similar within model types (e.g., within CreditRisk+ or within KMV) when the same version of the model is used. In some model types, the outputs are almost identical.
• Within model types, most differences in outputs reflect differences in the following factors: model inputs, pre-processing (i.e., packaging transactions into a readable format), valuation, errors in model usage during testing.
• Some differences in model outputs also can be attributed to differences in the analytical engines used in models and in versions of the same model. In particular, differences in the approach to valuation and correlation calculation methods were the key drivers of discrepancies in outputs among the publicly available models.
• With respect to portfolio construction, the most significant drivers of portfolio risk were credit quality (tested by subjecting portfolios to specified downgrade scenarios), correlation, and loss in event of default. Because the testing exercise adopted a uniform time horizon of one year for all portfolios, maturity effects were not systematically tested in this exercise.
• Most models permit significant flexibility in application, allowing users to set assumptions and parameters consistent with portfolio characteristics and risk appetite. This means that no two banks are likely to implement a credit risk model in an identical manner.
• The unique structure of certain portfolios (e.g., mortgages and emerging markets) currently require banks to use specially tailored modelling systems whose analytic engines, assumptions, and parameters may differ from those of other portfolios.
• Robust internal models exist to assess credit risk in mortgage and retail portfolios based more on scoring methodologies and aggregate measures of loss than on a direct assessment of default probabilities and migrations.
• Robust internal models designed to process available information in middle market portfolios exist and banks are beginning to apply some publicly available models in this area as well.
• Banks use credit risk models for different purposes. In addition, even the same model can generate different outputs if assumptions and parameters vary. Therefore, it is of paramount importance that risk managers understand why a model is generating a specific set of outputs. Variance in output across institutions using the same modelling system need not be alarming if the variance can be attributed to parameter differences generated by a bank's choices concerning portfolio construction and risk management priorities.
• A variety of methods currently exist for generating fair value estimates for credit transactions, even within individual models. Valuation methods, changes in spreads, discount rates, and the treatment of cash flows all can affect significantly model outputs. Legitimate differences in approach may exist. Therefore, risk managers should focus on ensuring that all such elements are consistent with the bank's strategic and tactical focus.
• Model choices and parameter settings are often driven by a bank's underlying risk management philosophy. Therefore, it is important to understand how model constructions vary and whether a model places emphasis on the risk factors deemed most important in a portfolio for purposes of managing and controlling risks in the portfolio.

The banks that participated in the IIF-ISDA Credit Risk Modelling Project were: Abbey National Bank PLC, ABN-AMRO Bank, Banco Santander Central Hispano, Banque Nationale de Paris, Banque Brussels Lambert, Bank One, Barclays Bank PLC, Chase Manhattan Bank, CIBC, Citibank, Credit Suisse, First Boston, Deutsche Bank AG, Dresdner Bank AG, FleetBoston Financial, HSBC, ING Bank, JP Morgan, KBC, Lloyds TSB, NatWest, Banque Paribas, Royal Bank Financial Group, Royal Bank of Scotland, Skandinaviska Enskilda Banken, UBS AG.

The Institute of International Finance (IIF) is the global association of financial institutions with more than 315 member organizations. The IIF has its head office in Washington D.C. The IIF web site is www.IIF.com.

Please click here to obtain an order form.