Srovnání logistické regrese a survival analýzy při odhadu pravděpodobnosti defaultu

The aim of this thesis is to compare the discriminatory power of selected models in predicting the probability of default over a twelve-month horizon. The first model, which represents the group of survival analysis models, is the Cox proportional hazards model. The key property of this model for the purposes of this thesis is the ability to estimate the probability of default over any given time horizon. The second model is logistic regression, which represents the industry standard and is always built for a specific horizon. For the model comparison, the ROC curves and AUC shall be used. On the selected dataset of mortgage loans, logistic regression proved to be the model with superior discriminatory power. The flexibility of the Cox model in estimating probabilities across arbitrary horizons does not, in this case, sufficiently compensate for its weaker discriminatory power.