基于分数阶差分ARFIMA模型及预测效果研究金秀.pdf

2007 9 Sep1, 2007 26 5 Application of Statistics andM anagement V ol126 No15 : 1002- 1566(2007) 05- 0896- 12 ARFIMA 金 秀 姚 瑾 庄新田 (, , 110004) : 采用MRS分析法对香港恒生指数周数据序列的长期记忆性进行 究, 并建立 ARFIMA 模 型, 推导了分数阶差分的计算过程对分数阶差分的 ARFIMA 模型与一阶差分的 ARFIMA模型进 行了比较, 发现应进行分数阶差分的序列, 简化成一阶差分后, 就有可能丢失许多有价值的信息, 导致建模误差增大进一步使用 ARFIMA模型预测公式进行预测, 结果显示 ARFIMA 模型预测效 果不理想在对香港恒生指数周数据进行预测时, ARFIMA 模型几乎是失效的,并从两 个不同的角 度论证了这一结果出现的必然性 : 分数阶差分; ARFIMA模型;长期记忆性;预测效果 : F830 : A Study of ARFIM A M odel and Its Forecast Perform an ce B ased on Fractional D ifferencing JIN Xiu, YAO Jin, Z UANG X inOtian ( SchoolofBusiness Administration, Northeastern University, Shenyang 110004, China) A bstrac t: The long- term memory of ongKong ang Sheng index using MRS analysis was studied, established ARFIMA model for it, and detailed the procedure of fractionaldifferencing. Furthermore, we compared theARFIMA modelbuiltby thismeanswith the one that took first order differencing as an alternative. The result showed that, if doing so, many useful information of tmi e serieswould be lost. The forecast formula ofARFIMA modelwas corrected according to the method of fractionaldifferencing, andwas employed in the empiricalstudy. Itwas illustrated that the forecast performance ofARFIMA modelwas not as not aswe expected since theARFIMA modelwas ineffective in forecasting ang Sheng index. The certainty of this conclusion was proposed from wt o different aspects. K ey w ord s: fractional difference; ARFIMA mode;l long- term memory; forecasting performance 0 , , Peters [ 1- 2] R/S Lo R/S