系统工程理论与实践
繫統工程理論與實踐
계통공정이론여실천
Systems Engineering—Theory & Practice
2013年
2期
296~307
,共null页
高频波动率 预测 自适应的不对称性HAR-CJ—D—FIGARCH模型 SPA检验
高頻波動率 預測 自適應的不對稱性HAR-CJ—D—FIGARCH模型 SPA檢驗
고빈파동솔 예측 자괄응적불대칭성HAR-CJ—D—FIGARCH모형 SPA검험
high-frequency volatility; forecast; adaptive asymmetry HAR-CJ-D-FIGARCH model; SPA test
在HAR-GARCH模型和HAR-CJ模型的基础上构建了自适应的不对称性HARCJ—D—FIGARCH模型，并用以对中国股市高频波动率进行了预测，然后利用上证综指2000年至2008年的高频数据实证检验了中国股市高频波动率的特征，最后运用SPA检验评价和比较了构建的模型与其他6类高频波动率模型的样本外预测能力．结果表明：中国股市高频波动率同时具有长记忆性、结构突变、不对称性和周内效应等特征；结构突变仅部分解释其长记忆性；高频波动率连续性成分的长记忆性很强，而跳跃性成分的长记忆性非常弱．相比于其他6类模型，自适应的不对称性HARCJ—D—FIGARCH模型对样本内数据的拟合效果最好，同时也是样本外预测性能最好的模型．
在HAR-GARCH模型和HAR-CJ模型的基礎上構建瞭自適應的不對稱性HARCJ—D—FIGARCH模型，併用以對中國股市高頻波動率進行瞭預測，然後利用上證綜指2000年至2008年的高頻數據實證檢驗瞭中國股市高頻波動率的特徵，最後運用SPA檢驗評價和比較瞭構建的模型與其他6類高頻波動率模型的樣本外預測能力．結果錶明：中國股市高頻波動率同時具有長記憶性、結構突變、不對稱性和週內效應等特徵；結構突變僅部分解釋其長記憶性；高頻波動率連續性成分的長記憶性很彊，而跳躍性成分的長記憶性非常弱．相比于其他6類模型，自適應的不對稱性HARCJ—D—FIGARCH模型對樣本內數據的擬閤效果最好，同時也是樣本外預測性能最好的模型．
재HAR-GARCH모형화HAR-CJ모형적기출상구건료자괄응적불대칭성HARCJ—D—FIGARCH모형，병용이대중국고시고빈파동솔진행료예측，연후이용상증종지2000년지2008년적고빈수거실증검험료중국고시고빈파동솔적특정，최후운용SPA검험평개화비교료구건적모형여기타6류고빈파동솔모형적양본외예측능력．결과표명：중국고시고빈파동솔동시구유장기억성、결구돌변、불대칭성화주내효응등특정；결구돌변부부분해석기장기억성；고빈파동솔련속성성분적장기억성흔강，이도약성성분적장기억성비상약．상비우기타6류모형，자괄응적불대칭성HARCJ—D—FIGARCH모형대양본내수거적의합효과최호，동시야시양본외예측성능최호적모형．
Based on HAR-GARCH model and HAR-CJ model, this article proposes an adaptive asym- metry HAR-D-CJ-FIGARCH model and utilizes it to conduct a volatility forecast. And then this article investigates the various properties of volatility simultaneously by utilizing the proposed model and the high-frequency data from SSEC from 2000 to 2008. Finally this article employs the SPA test to evaluate and compare the out-of-sample forecast performance of 7 high-frequency volatility models. The results show that the high-frequency volatility in Chinese stock markets has long term memory, structural breaks, asymmetry, and day-of-the-week effects. The structural breaks can only partially explain the long memory. The continuous components of high-frequency volatility have strong long-term memory, and the long-term memory of jump components is very weak. As compared to the other 6 models, the proposed model improves the in-sample fitting significantly, and provides the best out-of-sample forecast.
