Предсказание негативного события для фирмы-должника

Current work addresses the accuracy of credit events probability estimation and works on enhancing the existing probability of default models. Firstly, the hypothetically significant set of factors is outlined, followed by a method of bringing the classical expected loss formula closer to the existing default procedures. This is done via the introduction of an additional "payment default" category, which is defined to be better than bankruptcy but worse than the full financial safety. The research is logically separated into three Branches of models, depending on the aims of each model. Branch I deals with the enhacement of the existing probability of default models by including several new effects, which were previously omitted into the probabilisitic approach. Branch II outlines the basic aproach to the separation of the credit events into payment defaults and bankruptcies. It uses the sub-sample of problematic firms only. Branch III uses the natural instrument for the three-way ordered classification problem, which is the ordered logit model. Next, the work proposes the procedure of obtaining the probability estmates for the future status of the firm by two approaches. The first approach, called the chained approach, uses the consequent estimation by the optimal models from Branches I and II. The first model separates the safe state firms from all others, providing the probability of the adverse event. The second model gives the estimates of conditional probabilities for each of the adverse events given the problematic state of the firm. However, the ordered logit model computes them straightforwardly. In order to compare the chained approach with the ordered logit approach, the approaches are assessed on their ability to separate the safe firms from the others and segregate the problematic events. Finally, it was empirically demonstrated,that the chained approach is advantageous over the ordered logit model in scope of this paper.