Forecasting corporate defaults is a classic topic in finance. Default risk models are used in many different areas of research, including asset pricing, bank regulation or corporate valuation. The literature using these models is too large to be summarized in this blog post. Moreover, powerful default risk models are essential for lenders. Together with two co-authors, I revisit several popular models and test them in a new database of corporate defaults in the German stock market. Hence, we provide an out-of-sample test for these models. In general, research in finance is heavily dominated by US institutions and empirical research is almost entirely based on US data. Generalizing findings and applying models in different markets and jurisdictions often seems quite inappropriate. The economic dynamics of distress and bankruptcy are likely to be affected by the jurisdiction and other institutional factors (see for example Davydenko and Franks (2008)). Hence, out-of-sample tests for default risk models are necessary. The results are now available here.
Erroneously, it is a common belief that defaults are extremely rare in the stock market, which is dominated by large firms. We find that corporate defaults have, at least since the end of the 1990s, been a characteristic feature of the German stock market. Yearly default rates of up to 3.6% illustrate the practical use of the models tested in our article. Specifically, we test the forecasting performance of the Merton (1974) distance-to-default model, the Altman (1986) Z-Score and the Campbell (2008) failure score. Broadly speaking, default risk models are based on two different methodologies. Structural models use economic theories to extract default risk indicators from security prices. The prime example for such a model is the Merton (1974) distance-to-default. This model assesses the likelihood of the firm’s asset value dropping below the value of its debt, which would theoretically render the firm worthless. The second class of models is based on a purely empirical approach and uses regressions of firm characteristics on default indicators to compute probabilities of default. Hence, these models are generally called statistical models. A classic example for such a model is the Altman (1968) Z-Score, which uses several balance sheet indicators, for example leverage ratios, as exogenous variables. A few years ago, Campbell et al (2008) proposed another model, which they call failure score. This model uses information about firm valuation as exogenous variables in addition to the traditional balance sheet indicators.
The bottom line of our findings is that the Campbell (2008) failure score performs best in the German stock market. Our new results confirm previous findings, which show that the contribution of traditional balance sheet and profitability variables to the forecasting performance is relatively minuscule (Shumway (2001)). The most influential forecasting variables are derived from market information. Furthermore, we also assess the impact of the different default risk models on the lending business in a simulated loan market. The results confirm that, compared to the other models, the failure score leads to better credit decisions and higher bank profitability.
Richard L. Mertens is a Researcher at the Faculty of Business Studies & Economics of the University of Bremen.