金融高频数据跳跃波动研究——基于大数据核函数支持向量机的方法  

Study on Jump Volatility of Financial High-frequency Data:Based on the Method of Big-data Kernel Functions SVM

作　　者：柳向东[1] 李文健 LIU Xiang-dong;LI Wen-jian(School of Economics,Jinan University,Guangzhou 510632,China)

机构地区：[1]暨南大学经济学院,广东广州510632

出　　处：《统计与信息论坛》2018年第9期23-30,共8页Journal of Statistics and Information

基　　金：国家自然科学基金资助项目<带Lévy跳的多因子市道轮换框架下的仿射利率结构模型>(71471075);教育部人文社会科学研究一般项目<基于市道轮换框架下带Lévy跳的高频数据的波动率>(14YJAZH052);中央高校基本科研业务费专项资金项目"暨南跨越计划"(15JNKY003)

摘　　要：高频数据价格波动率具有明显的长记忆性特征和"尖峰厚尾"现象,运用沪深300指数5分钟高频数据,通过已实现波动率和已实现双幂次变差对资产价格的连续性波动和跳跃波动进行建模,得到进行波动率短期预测的HAR-RV模型、HAR-lnRV模型及HAR-JV-CV模型。将预测效果较好的模型与不同核函数下的支持向量机相结合,结果表明:不同核函数之间存在较大的相似性;将跳跃波动预测模型与支持向量机相结合能够提高模型的短期预测精度。波动率的预测对于政府监管部门、投资者和资本市场来说均具有重要意义。The price volatility of high-frequency data has the characteristics of obvious long memory and the"leptokurtosis and fat-tail distribution".Based on 5 minutes high-frequency data of CSI 300 index,this paper builds a model of continuous volatility and jump volatility of asset price according to the realized volatility(RV)and the realized bi-power volatility(RBV),and obtains the HAR-RV model,the HAR-lnRV model and the HAR-JV-CV model which are used to give short-term prediction of volatility.Then it combines the model which provides pretty prediction with the support vector machine under different kernel functions.The result shows that there are great similarities among different kernel functions;and combining the jump volatility prediction models with the support vector machine could improve the short-term prediction accuracy of the model.The prediction of volatility is of great significance for the government regulators,investors and capital markets.

关 键 词：已实现波动率 HAR-lnRV模型 支持向量机 短期预测 核函数 

分 类 号：F224[经济管理—国民经济] F830