BAYESIAN WAVELET SHRINKAGE BASED ON CHAMPERNOWNE PRIOR WITH APPLICATIONS

Abstract

Bayesian wavelet shrinkage have been widely used in several areas to reduce noise in data analysis. In this work, we propose a mixture of a Dirac function with the Champernowne distribution as prior probabilistic distribution to wavelet coefficients in a nonparametric regression problem. The associated bayesian shrinkage rule has parameters that are easily interpreted and its performance in simulation studies was superior in most of the considered scenarios against methods available in the literature and used for comparison. Applications of the method to  neuronal action potentials and the São Paulo Stock Market Index (IBOVESPA) datasets are made.