Recovery Rates for Basel II and Bank-Wide Risk Management

SKS Advisory has already repeatedly been able to provide extensive support in the development and implementation of LGD models to banks based in Germany and with international operations. On the basis of this experience and our long-standing expertise, we are able to offer our clients unique consulting, development and implementation services as regards this particular regulatory topic.

360° Experience in LGD modelling

SKS has already been able to successfully develop and implement BaFin-certified LGD models for all the business divisions shown in the graphic below. By way of illustration, the "traffic light" dots denote the degree of difficulty that can be expected in the development of the model, based on experience, as well as the general level of knowledge about LGD in this area.



LGD in bank-wide risk management

The German Solvency Regulation (Solvabilitätsverordnung, "SolvV") prescribes the mandatory use of internal LGD models, which have to be certified for regulatory purposes. For institutions this constitutes an important step in the direction of consistent bank-wide risk management.


When compared with traditional models with estimates based purely on expert opinion, LGD models created on the basis of statistical methods and then optimised by experts have a strengths and weaknesses profile that is complementary (please see diagram below). This makes LGD models a valuable bank-wide risk management instrument.



The SKS generic data repository enables a quick start

On the basis of extensive experience in projects for different banks, SKS has created a generic data repository, which includes all the LGD modelling requirements. When an implementation is carried out, this data repository is adapted specifically for the needs of the contracting bank. Such an approach means that it is possible to design the data repository, complete the definition of loss as well as to commence the implementation phase within a period of around one month.



This diagram gives a very rough overview of the classes involved and does not show any attributes. The modelling is done on the basis of Unified Modelling Language (UML 2.0), which can also be used directly for generating code or a database schema.

Empirically based LGD for states and regional administrative agencies

Despite difficulties with respect to data records, it is possible to develop empirically based LGD models that not only fulfil the strict requirements of the "SolvV" but also satisfy experienced credit experts. With SKS you can retrieve external data sources and combine them with your internal data to create the basis for valid and empirically based LGD modelling. The LGD ratios for all bank internal historic sovereign defaults have to be determined on the basis of the overall economic loss in relation to the exposure at the time of default.

 

Although, here, as is usual with other portfolios, there are already considerable difficulties; in view of the definitions of economic loss and default that are deemed to be acceptable by the regulator, a great deal of effort is required for the evaluation of historic internal databases. Furthermore, there are specific complications that have to be taken into account with respect to establishing default criteria for sovereigns.

The key challenge in LGD modelling for sovereigns is, however, the very low number of default events in internal portfolios.

Therefore, institutions have to ask themselves how they can supplement their internal data with external data in a way that will be cost-effective as well as technically and also statistically valid. If the number of internal default events is too low then an additional evaluation of published historic data on sovereign defaults, or restructuring agreements (sources, e.g. IIF, World Bank) will be required.Furthermore, other valuable information can be gained from methods that are based on historic or current market data about bonds or CDS. In the course of this, all the data sources have to be integrated into one single data source in a way that is controlled and expert-driven as well as statistically valid (e.g. based on the Regularized Maximum Likelihood approach from Friedman and Sandow). Finally, there is a large amount of benchmark information that also has to be systematically assigned to an overall framework. Moreover, in the last few years, research into the causes of and trigger mechanisms for sovereign defaults has been able to provide interesting and important inputs.