Despite a surge in the research efforts put into modelling credit risk during the past decade, few studies have incorporated the impact that macroeconomic conditions have on business defaults. In this paper, we estimate a duration model to explain the survival time to default for counterparts in the business loan portfolio of a major Swedish bank over the period 1994-2000. The model takes both firm specific characteristics, such as accounting ratios and payment behaviour, loan related information, and the macroeconomic stance into account. The output gap, the yield curve and consumers’ expectations of future economic development have significant explanatory power for the default risk of firms. We also derive two measures of portfolio credit risk, Value-at-Risk-type and Expected Shortfall, by means of a Monte-Carlo simulation method that employs model-based probabilities of default. This two-step approach allows us to (i) make individual forecasts of default risk conditional on firm and loan characteristics, as well as prevailing macroeconomic conditions, (ii) study the evolution of portfolio credit risk over time while conditioning on the macroeconomy. We also compare our model with a frequently used model of firm default risk that conditions only on firm-specific information. The comparison shows that while the latter model can make a reasonably accurate ranking of firms’ according to default risk, our model, by taking macro conditions into account, is also able to pin down the absolute level of (default, and thus also credit) risk.
The literature on credit risk modeling is extensive and growing. A pioneering contribution from the 1960’s is Altman’s study of business default risk [3]. Following Altman, a number of authors have estimated various types of default risk models on cross-sectional data sets. For example Altman [4] [5], Frydman, Altman and Kao [16], Li [28], and Shumway [34], who all focus on the analysis of bankruptcy risk at the firm level.
In the last decade, a whole range of modeling techniques has been developed to analyze portfolio credit risk. Broadly viewed, there are four groups of portfolio credit risk models. The first group is ’structural’ and based on Merton’s [32] model of firm capital structure: individual firms default when their assets’ value fall below the value of their liabilities. Examples of such a microeconomic causal model are Credit Metrics and KMV’s Portfolio Manager. The second group consists of econometric factor risk models, like McKinsey’s Credit Portfolio View. McKinsey’s model is basically a logistic model where default risk in ’homogeneous’ subgroups is determined by a macroeconomic index and a number of idiosyncratic factors. These models employ closely related Monte Carlo simulation approaches to calculate portfolio risk. Both groups consist of ’bottom-up’ models that compute default rates at either the individual firm level or at sub-portfolio level. Both thus require a similar kind of aggregation. The third group contains ’top-down’ actuarial models, like Credit Suisse’s Credit Risk+, that make no assumptions regarding causality. Finally, a number of authors, such as Carey [9], use non-parametric methods.
Koyluoglu and Hickman [25] provide an elaborate description of the above mentioned types of portfolio credit risk models. They note that all model types, despite their differences, are built on three more or less general components to calculate portfolio loss distributions. First, they contain some process that generates conditional default rates for each borrower in each state of nature and a measure of covariation between borrowers in different states of nature. Second, their set-up allows for the calculation of conditional default rate distributions for sets of homogeneous sub-portfolios (e.g., rating classes) as if individual borrower defaults are independent, since all joint behavior is accounted for in generating conditional default rates. Third, unconditional portfolio default distributions are obtained by aggregating homogeneous sub-portfolios’ conditional distributions in each state of nature; then conditional distributions are averaged using the probability of a state of nature as the weighting factor. In Sections 3 and 4 we will see that the duration model approach followed in this paper is contained by this general description of portfolio credit risk models.
Download
Corporate Credit Risk Modelling and the Macroeconomy
