B1136
Title: Long range dependence, fractional renewal models and Bayesian inference
Authors: Nicholas Watkins - London School of Economics and Political Science (United Kingdom) [presenting]
Christian Franzke - University of Hamburg (Germany)
Abstract: Since the 1960s, long range dependence (LRD) as embodied by the fractional Gaussian noise and ARFIMA models, has been a well-studied mechanism for the origin of $1/f$ noise and the Hurst effect. Two new avenues of research will be discussed. The first concerns breakpoints. These have long been known to be a source of the Hurst effect, but recent research by one of us has shown that Mandelbrot had proposed a model with power law intervals between the breaks as early as 1963, and that by 1965-67 he was showing how this was an alternative non-ergodic model for $1/f$ noise, with consequences for model choice and time series interpretation that are increasingly becoming topical in physics and elsewhere. The second concerns Bayesian inference when an LRD model is plausible. We will discuss our recent work on a novel systematic Bayesian approach for joint inference of the memory and tail parameters in an ARFIMA model with heavy-tailed innovations.
Journal