應用基因演算法決定SETAR門檻區間及其應用

Abstract

近年來，面對傳統線性時間序列的預測問題，有許多技術上的改良而被大量廣泛的使用，但是線性模式往往無法處理常常發生結構改變(structural changes)的問題，這使得非線性(nonlinearity)時間序列轉折點的研究越來越受到重視，利用非線性時間序列解決實例更可以貼近真實情況。再者，隨著模糊理論的蓬勃發展以及區間軟計算(soft computing)的成熟，相較於點估計預測方法所需的嚴格假設，區間估計方法的假設寬鬆許多並且能符合實際情況，可以提供給決策者更彈性的選擇。本文將應用基因演算法(genetic algorithms)針對模糊區間資料(fuzzy data)作模糊分析(fuzzy analysis)，找出資料轉折的門檻區間(threshold interval)，藉此發展出非線性的區間門檻自迴歸模式(interval SETAR model)，最後以台股為例，建構出門檻自迴歸模型與傳統區間ARIMA模式比較，藉此探討其預測方法的效率評估與準確性。

In recent years, in the face of traditional linear time series forecasting problems, there are many technical improvements and widely used. But linear model are often unable to deal with the problem often happens structural changes, which makes the nonlinear turning point for the study of the time series more and more attention. Use nonlinear time series more close to the real situation. Moreover, with the fuzzy theories flourish and soft computing mature, compared to the point estimate methods required strict assumptions, interval estimation method which without many assumptions can meet the actual situation. It can be provided to decision-makers more flexibility of choice. In this paper, the application of genetic algorithms for fuzzy data to identify structural changes interval (threshold interval), so as to develop the nonlinear range threshold autoregressive mode (interval SETAR model), and finally, for example, the Taiwan stock market, construct a threshold autoregression model with the traditional interval ARIMA model to investigate the prediction method efficiency and accuracy.