ARFIMA模型在金融時間序列的應用
首發時間:2018-10-31
摘要:本文系統地討論了如何對分整自回歸移動平均(Autoregressive fractionally integrated moving average, ARFIMA)模型進行參數估計以及相應的建模。具體來說,三種方法ADF、PP和KPSS三種方法被用來檢驗上證指數和深證成指兩條序列的平穩性,并用經典的R/S分析法和修正的R/S分析法以及V/S分析法來分析兩條序列的長記憶性。結論顯示一致支持上證和深證的日收益序列都有長記憶性,而且上證序列的長記憶性比深證的強?;陂L記憶性的檢驗結果,本文對兩條序列采用B-J 方法進行ARFIMA建模。通過信息準則比較確定為,ARFIMA$(6,0.1520,2)$是描述上證日收益序列長記憶性最合適的模型,ARFIMA$(5,0.1282,2)$是描述深證成指日收益序列長記憶性最合適的模型。通過實證研究得出,我國股市中確實存在長記憶性,但中國股市缺乏有效性。
For information in English, please click here
Application of ARFIMA model in financial time series
Abstract:This paper systematically discusses the parameter estimation and modeling of the integral autoregressive moving average (ARFIMA) model in real-life case studies. ADF, PP, and KPSS are first employed to test the stationarity of the Shanghai and Shenzhen composite index. Then long-range correlation is measured for these two series using the R/S, modified R/S, and V/S analysis. The results demonstrates that the daily revenues sequences of Shangzhen and Shenzhen have long-range correlation but the former's is stronger. Based on the estimated long-range correlation, the B-J method is employed to construct the ARFIMA model for two series. ARFIMA(6, 0.1520, 2) and ARFIMA(5, 0.1282, 2) are specified as the best model to describe the Shangzhen and Shenzhen daily revenue series, based on the comparison of information criteria. It can be concluded from this case study there is indeed long-range correlation in China's stock market, which invalidates the efficient markets hypothesis in China.
Keywords: long range correlation;ARFIMA model;R/S analysis method.
引用
No.****
動態公開評議
共計0人參與
勘誤表
ARFIMA模型在金融時間序列的應用
評論
全部評論