Author:
- Marc Aurele Divernois
Abstract:
This paper proposes a machine learning approach to estimate physical forward default intensities. Default probabilities are computed using artificial neural networks to estimate the intensities of the inhomogeneous Poisson processes governing default process. The major contribution to previous literature is to allow the estimation of non-linear forward intensities by using neural networks instead of classical maximum likelihood estimation. The model specification allows an easy replication of previous literature using linear assumption and shows the improvement that can be achieved.
SSRN: