A new unsupervised threshold determination for hybrid models
Nehla Debbabi, Marie F. Kratz · 2014
A Gauss-GPD hybrid model that links a Gaussian distribution to a Generalized Pareto Distribution (GPD) is considered for asymmetric heavy tailed data. The paper proposes a new un-supervised iterative algorithm to find successively the junction point between the two distributions and to estimate the hybrid model parameters. Simulation results show that this method provides a reliable position for the junction point, as well as an accurate estimation of the GPD parameters, which improves results when compared with other methods. Another advantage of this approach is that it can be adapted to any hybrid model.