Estimation du risque de défaut par une modélisation stochastique du bilan : Application à des firmes industrielles françaises

An alternative method is proposed to estimate corporate default risk, beside multivariate models such as multiple discriminant analysis or neural networks. We use an Asset Liability Management method of risk estimation. This method is based on corporate bond valuation models. Whereas ALM is quite exclusively used by banks and insurance companies, we assess probabilities of default for French industrial firms. Following Janssen [1992], we suppose that the dynamics for the total assets and the total liabilities can be described by geometric Brownian motions. The probability of insolvency -i.e. the probability that net worth is negative - is then estimated and analysed as the probability of default. A repartition of the firms into two groups is done from the computed probabilities. A threshold is chosen ; any firm whose probability is lower than the threshold is classified in the group of the non-failed companies. Any firm whose probability is higher than the threshold is classified in the group of the failed companies. Rate of correct classification is assessed from bootstrap samples and compared to other business failure prediction models. The assessed probabilities discriminate the firms that filed for bankruptcy from the healthy firms one year but also two and three years prior to failure. Our study provides a simple and accurate indicator of corporate bankruptcy risk and proves that empirical applications of stochastic calculus to industrial firms allow to obtain good results.