Estimation du paramètre de multifractalité : régression linéaire, maximum de vraisemblance ou inférence Bayésienne ?

Abstract

Multifractal has nowadays become a standard tool in modern signal and image processing, mostly used to characterize scale-free spatial or temporal dynamics. While multifractality parameter estimation can be performed using well-documented procedures, it remains a challenging task for small sample-size observations. This works studies Expectation-Maximization estimators for the multifractality parameter and compares, by Monte Carlo simulations, their performance against state-of-the-art estimators for univariate small sample size time series.