CFE 2020: Start Registration
View Submission - CMStatistics
B1110
Title: De-biased Whittle likelihood for time series and random fields Authors:  Arthur Guillaumin - New York University (United States) [presenting]
Adam Sykulski - Lancaster University (United Kingdom)
Sofia Olhede - EPFL (Switzerland)
Abstract: Maximum likelihood parameter estimation of time series and spatial models is often intractable for non-trivial covariance structures. We will discuss the de-biased Whittle Likelihood - a method we recently developed that can estimate time-series parameters for massive datasets using Fast Fourier transforms. The procedure is related to, but distinct from, the standard and well-known Whittle Likelihood. We will make these distinctions more clear, both from a practical and theoretical point of view. We have adapted the procedure to allow for missing data through a framework based on that of modulated time series. We have found numerous application benefits, and I will showcase these through an application to the study of Venus' topography as well as simulation studies for two and three-dimensional spatial data.