Аннотация:The recent default of the multinational giants Enron, Parmalat and Worldcom clearly showed how accounting data can be misleading and far away from the true financial situation of a company. When financial fraud takes place, the models that use accounting data to predict default probabilities cannot be used since their forecasts are completely unreliable. To avoid such problems, we propose a novel approach that uses stock prices only, and allows to model departures from normality in stock returns dynamics, too. The parametric bootstrap, based on a conditional marginal model, is used to estimate the distribution of these estimated probabilities and to construct confidence bands. We show an empirical example with quoted Russian stocks as well as with American, Italian and Russian defaulted stocks, whose financial statements were found to be irregular.