The Shadow Rating Approach: Experience from Banking Practice

Abstract

In this article we will report on some aspects of the development of shadow rating systems found to be important when re-devising the rating system for large cor-porations of KfW Bankengruppe (KfW banking group). The article focuses on general methodological issues and does not necessarily describe how these issues are dealt with by KfW Bankengruppe. Moreover, due to confidentiality we do not report estimation results that have been derived. In this introductory section we want to describe briefly the basic idea of the shadow rating approach (SRA), then summarise the typical steps of SRA rating development and finally set out the scope of this article. The shadow rating approach is typically employed when default data are rare and external ratings from the three major rating agencies (Standard & Poor's, Moody's or Fitch) are available for a significant and representative part of the portfolio. As with other approaches to the development of rating systems, the first modelling step is to identify risk factors – such as balance sheet ratios or qualitative information about a company – that are supposed to be good predictors of future defaults. The SRA's objective is to choose and weight the risk factors in such a way as to mimic external ratings as closely as possible when there is insufficient data to build an explicit default prediction model (the latter type of model is e.g. described in Chap. 1. To make the resulting rating function usable for the bank's internal risk management as well as for regulatory capital calculation, the external rating grades (AAA, AA, etc.) have to be calibrated, i.e., a probability of default (PD) has to be attached to them. With these PDs, the external grades can then be mapped to the bank's internal rating scale., # Springer-Verlag Berlin Heidelberg 2011 37 The following modular architecture is typical for SRA but also for other types of rating systems: 1. Statistical model 2. Expert-guided adjustments 3. Corporate group influences/Sovereign support 4. Override The statistical model constitutes the basis of the rating system and will most likely include balance sheet ratios, macroeconomic variables (such as country ratings or business cycle indicators) and qualitative information about the company (such as quality of management or the company's competitive position). The statistical model will be estimated from empirical data that bring together compa-nies' risk factors on the one hand and their external ratings on the other hand. The model is set up to predict external ratings – more precisely, external PDs – as efficiently as possible from the selected risk factors. The second modelling layer of the rating system, that we have termed " Expert-guided adjustments " will typically include risk factors for which either no historical information is available or for which the influence on external ratings is difficult to estimate empirically. 1 Consequently, these risk factors will enter the model in the form of adjustments that are not estimated empirically but that are determined by credit experts. The third modelling layer will take into account the corporate group to which the company belongs or probably some kind of government support. 2 This is typically done by rating both the obligor on a standalone basis and the entity that is supposed to influence the obligor's rating. Both ratings are then aggregated into the obligor's overall rating where the aggregation mechanism will depend on the degree of influence that the corporate group/sovereign support are assessed to have. Finally, the rating analyst will have the ability to override the results as derived by steps 1–3 if she thinks that – due to very specific circumstances – the rating system does not produce appropriate results for a particular obligor. This article will focus on the development of the rating system's first module, the statistical model. 3 The major steps in the development of the statistical model are: 1