A0479
Title: Generalising dynamic semiparametric averaging forecasting for time series with discrete-valued response
Authors: Rong Peng - University of Southampton (United Kingdom)
Zudi Lu - University of Southampton (United Kingdom)
Fangsheng Ge - University of Southampton (United Kingdom) [presenting]
Abstract: The aim is to explore how to utilise the useful high-dimensional lagged information for dynamic forecasting of time series data with a discrete-valued response. The approach will generalise the existing flexible semiparametric marginal regression model averaging (MARMA) forecasting, a useful data-driven method designed for nonlinear forecasting of continuous-valued time series by the least squares averaging. A generalised MARMA (GMARMA) procedure has been suggested under a general time series exponential family of distributions, which flexibly accommodates nonlinear forecasting of discrete-valued response. Further, it allows lagged effects, including discrete-valued information for forecasting. A conditional likelihood model averaging method, instead of the least squares, is developed to estimate the average weights in the GMARMA under a beta-mixing time series data generating process with established asymptotic normality. Furthermore, an adaptively penalised GMARMA (PGMARMA) is suggested to select the important variables for improved forecasting. The oracle properties of the PGMARMA weights are established as if the true non-zero weights were known. These procedures are further supported by Monte Carlo simulations and empirical applications to forecasting the FTSE 100 index market moving direction and the UK road casualty data, which outperform many popular machine learning tools.
