．單位首頁
．系所簡介
．系主任
．強力師資
．研究所
．課程概況
．各類學程
．期刊論文
．招生資訊
．演講專區
．榮譽榜
．系學會
．系友天地
．就業訊息
96年12月28日(五) 2:00 ~3:00 P.M.
演講者姓名： 劉峰旗 博士候選人
演講者服務單位： 逢甲大學應用統計研究所
Threshold Autoregressive Moving-Average Models with Orders Selection
Abstract
Orders selection of a non-linear time series model is essential for time series modeling. This paper focuses on the orders selection of a threshold autoregressive moving-average model. It increases the complexity of model selection by considering a threshold model with unknown thresholds and delay lag of the threshold variable. The existent methods based on the information-theoretic criteria are hard to implement with a huge number of possible models and require setting of the delay lag and threshold in advance. The proposed method, which uses the idea of stochastic variable search, treats the uncertainties of the delay lag and threshold and simultaneously selects the correct orders of the threshold autoregressive moving-average model. The results of simulation study demonstrate the accuracy of the proposed method with an appropriate setting of hyper-parameters. Finally, the applications of real data sets show that the proposed method can provide the possible models with higher probabilities for our model choices.
Keywords: Bayesian analysis; mixture normal distribution; stochastic search variable selection method; subset selection; threshold models
