- 期刊
- OpenAccess
A Study on Automatically Selecting Subset Autoregressive Time Series Models
部份自迴歸時間序列模式自動化之研究
摘要
ARMA (p,q) model have been shown to be powerful class of models which are being widely used for forecasting. However, the model building approaches propossed by Box and Jenkins (1976) are cumbersome. Meanwhile, the identification and analysis of the seasonal time series is not quite easy. These disadvantages might limit there use in practice. Recently, a wide class of subset autoregressive models, denoted by SAR in brief, has been proposed by several researchers. However, the estimation of subset autoregressive time series has been a difficult problem because of the large number of possible alternative models involved. This problem of identifying an appropriate subset autoregressive model to represent a time series Z(t), which may be a seasonal one, is considered. A diagnostic check for this model proposed by Hokstad (1983) using the estimated cross correlation function between the observed series and the residuals is modified and investigated in this study. This method is studied by thousands synthetic data as well as several real data. The results indicate that this method can successfully detect the true model for a given time series.
關鍵字
無資料並列摘要
ARMA（p,q）模式廣泛被運用於預測上，同時證明爲一族十分有效之模式，然而，Box及Jenkins氏所提模式之建立方法過份繁瑣，對於具週期性之時間序列模式之辨認及分析亦相當不易，這些缺點限制其在實際上之應用。近年來，幾個研究者提出一族部份自迴歸模式，然而，由於有許多可能之代替模式，其在估計上困難較大，所以，本研究對如何辨認一可行部份自廻歸模式（可能具週期性）之時間序列問題進行探討。Hokstad提議利用實測值與誤差值之交叉相關函數（cross-correlation function）來檢測模式，本研究將其修正並探討其可行性。數千個合成資料與實測資料用來測試本研究所提之方法。結果顯示本方法可十分成功地測知某一時間序列之正確模式。