You are here: Home / Seminars / 13/08/2021 - Dynamic Bayesian models for non-linear non-stationary time series processes - Speaker: Alvaro Faria (Open University, UK)

13/08/2021 - Dynamic Bayesian models for non-linear non-stationary time series processes - Speaker: Alvaro Faria (Open University, UK)

When Aug 13, 2021
from 02:00 PM to 04:00 PM
Add event to calendar vCal
iCal
UFSCar/USP joint Seminar
Scheduled for:
Aug 13, 2021, at 2:00 pm
(GMT-03:00) Brasilia Standard Time - Sao Paulo
  
 
Speaker:
Alvaro Faria (Open University, UK)
 
Title:
Dynamic Bayesian models for non-linear non-stationary time series processes
 
Abstract: 

Dynamic Bayesian Smooth Transition Autoregressive (DBSTAR) models are proposed for non-linear autoregressive time series processes as alternative to both the classical Smooth Transition Autoregressive (STAR) models of Chan and Tong (1986) and the Bayesian Simulation STAR (BSTAR) models of Lopes and Salazar (2005). Unlike the STAR and BSTAR models, DBSTAR models are sequential polynomial dynamic analytical models suitable for inherently non-stationary time series with non-linear characteristics such as asymmetric cycles. As they are analytical, they also avoid potential computational problems associated with BSTAR models and allow fast sequential estimation of parameters. Two types of DBSTAR models, namely the Taylor and the B-splines DBSTAR models are formulated. A harmonic version of those models, that accounted for the cyclical component explicitly in a flexible yet parsimonious way, were applied to the well-known series of annual Canadian lynx trappings and showed improved fitting when compared to both the classical STAR and the BSTAR models. Another application to a long series of hourly electricity loading in southern Brazil, covering the period of the South-African Football World Cup in June 2010, illustrates the short-term forecasting accuracy of fast computing harmonic DBSTAR models that account for various characteristics such as periodic behaviour (both within-the-day and within-the-week) and average temperature.


Bio:

Alvaro Faria is a lecturer in Statistics at the School of Mathematics and Statistics of the Open University, UK. He was previously a lecturer in Management Science at Lancaster University, and a research fellow in Statistics at Warwick University. He has a PhD in Statistics from Warwick, an MSc and a BSC in Systems Engineering from PUC-Rio. His current research interests include Bayesian time series forecasting, the combination of statistical models, expert judgement, Bayesian modelling of space-time processes and the application of forecasting models in tailings dams.

Filed under: