Ebook Credit Cycles and Macro Fundamentals
Systematic credit risk factors play a dominant role in current credit risk management. Traditionally, credit scoring methodologies focus on assessing the credit risk of individual counterparties (see i.a. Altman (1983), Altman (2000)). Though important, at the portfolio level most of the idiosyncratic risks can be diversified and only the systematic credit risk components remain, see, e.g., Lucas et al. (2002), Schönbucher (2001), Frey and McNeil (2003). This also holds if bond or loan portfolios are repackaged into new products like CDOs. In order to assess the credit risk at the portfolio level, it is important to model the correct dynamics of systematic credit risk components.
In this paper, we use the methodology of Koopman et al. (2005) to estimate the credit cycle directly from rating and default data at the micro level using intensity models with latent common risk factors. The data used are rating and default transitions for U.S. corporates rated by Standard and Poor’s and observed over December 1980 to June 2005. We condition the credit cycle on a number of macro economic fundamentals, reflecting the state of the business cycle, bank lending conditions, and financial market conditions. In line with results by for example Couderc and Renault (2005), these variables appear to capture part of the credit cycle dynamics. The models, however, turn out to be dynamically mis-specified as there is strong remaining autocorrelation in the intensities. If we account for this, the significance of many of the macro variables disappears. The results are robust to a variety of specifications of the model. This leaves us with the puzzle of what macro fundamentals drive default and (re-)rating behavior.
The formal testing procedure for dynamic mis-specification introduced in this paper constitutes a powerful tool in the empirical modeling of intensities. For example, in the current paper we model intensities of rating and default transitions on both observed macro fundamentals and on an unobserved credit cycle component. If the fundamentals explain the credit cycle completely, the unobserved component should drop from the analysis. This can be tested using standard likelihood ratio tests. The computation of the likelihood ratio test, however, is not trivial because the latent component must be integrated out of the likelihood. We describe the importance sampling methodology that makes these types of tests possible.
Empirically, the relation between default rates and growth has been addressed in a number of studies. Fama (1986) and Wilson (1997), regress default rates on observed macro variables and find cyclicality in probabilities of default (PDs), particularly in the case of economic down turns when PDs increase significantly. Koopman and Lucas (2005) concentrate on the time series dimension of PDs and present evidence of co-cyclicality between GDP and default rates. Kavvathas (2001) shows the influence of the term structure of interest rates over the rating migration (including default) intensities using parametric and semi-parametric duration models. Carling et al. (2002) employ the semi-parametric duration model of Cox (1972) conditioning on both firm specific and macroeconomic variables to analyze a dataset on business loans.
Duffie, Saita and Wang (2006) also incorporate firm specific information in their duration models. Their results indicate that both the level of real economic activity and the term structure of interest rates are important determinants of default risk. A commonality between the papers based on the intensity framework for credit risk is that hardly any attention is paid to the correct specification of the model. This is particularly relevant given the demonstrated stickiness of aggregate rating migrations and defaults.
Posted in :
