Time Series Forecast Using AR-Belief Approach

Nantiworn Thianpaen, Jianxu Liu, Songsak Sriboonchitta

Abstract

This paper aims atapplying a recent new approach to predicting thegrowth rate of ThailandGDP. The new approach will provide uncertainty aboutpredicted values solelyfrom observed data, without the need to supply some subjective priordistribution on unknown model parameters. This is achieved bybuilding a belief function(i.e., a distribution of a random set) from the likelihoodfunction given theobserved data, and use it to assess prediction uncertainty. Withour sampling model as anautoregressive time series model, we demonstrate empirically that this approachcan provide a reliable confidence interval for predicted values.