A0207
Title: Nonparametric estimation of splicing points in cost distributions through a transformation-based approach
Authors: Benedikt Funke - Cologne University of Applied Sciences (Germany) [presenting]
Masayuki Hirukawa - Ryukoku University (Japan)
Abstract: The tail of a distribution beyond some threshold is of importance and interest in academics and industries. The aim is to investigate the nonparametric estimation of a threshold in distributions of nonnegative cost variables such as incomes or insurance payments. It starts by interpreting the threshold as a jump location or splicing point in a distribution. Since the threshold typically lies in the right-tail region, it is often difficult to estimate it due to a small jump size and/or sparseness of the data. Then, the proposal is to transform the original nonnegative observations onto the unit interval and detect the threshold in the transformed scale using the asymmetric beta kernel. The data transformation makes the threshold estimation easier because the jump size is magnified, and two adjacent observations become closer to the unit interval. It is demonstrated that the threshold estimator is consistent with a faster convergence rate than the parametric one and asymptotically normal when suitably implemented. Monte Carlo simulations and real data examples illustrate attractive properties and practical relevance of the proposal in several different use cases.
