A0372
Title: Non-linearity and the distribution of market-based loss rates
Authors: Matthias Nagl - Universtiy of Regensburg (Germany)
Daniel Roesch - University of Regensburg (Germany)
Abstract: The extended linear beta regression is synthesized with a neural network structure to model and predict the mean and precision of market-based loss rates. Non-linearity in mean and precision is incorporated in a flexible way and the problem of specifying the underlying form in advance is resolved. As a novelty, it is shown that the proportion of non-linearity for the mean estimates is 14.10\% and 80.37\% for the precision estimates. This implies that especially the shape of the loss rate distribution entails a large amount of non-linearity. Furthermore, trainable activation functions are derived to allow a data-driven estimation of their shape. This is important if predictions have to be in a certain interval, e.g., $(0; 1)$ or $(0;1)$. It is shown how the new methods can be used for management decisions by conducting a scenario analysis. It is found that the estimated distributions are more refined compared to traditional models which can help financial institutions to better identify different risk profiles across their creditors and in different macroeconomic states.
