基于遗传算法的预测结合方法及其应用  

Prediction Combination Method Based on Genetic Algorithm and Its Application

作　　者：瞿慧[1] 王子旭 Qu Hui;Wang Zixu(School of Management and Engineering,Nanjing University,Nanjing 210093,China)

机构地区：[1]南京大学工程管理学院

出　　处：《统计与决策》2019年第23期67-70,共4页Statistics & Decision

基　　金：国家自然科学基金资助项目(71671084)

摘　　要：金融资产的方差与协方差是资产配置、套期保值、风险管理等实务应用的关键参数,因此对多元波动率预测模型的研究具有重要意义。文章采用预测结合技术,根据设定的优化准则对多个多元波动率模型的预测加权结合作为协方差的预测,其中最优权重通过遗传算法确定并动态调整。以常用低频与高频多元波动率模型为待选模型,沪深300指数与股指期货为实证数据,在多种常用损失函数下的样本外预测性能比较指出,基于遗传算法的线性预测结合在指数变化平稳与震荡期间均可以获得统计上显著较高的预测精度,是较使用单个多元波动率模型更稳健的选择。Variance and covariance of financial assets are the key parameters of asset allocation,hedging,risk management and other practical applications,thus it is of great significance to study multiple volatility forecasting model.In this paper,according to the set optimization criteria,the prediction weighted combination of multiple multivariate volatility models is used as the prediction of covariance,in which the optimal weight is determined and adjusted dynamically by genetic algorithm.Taking the commonly used low frequency and high frequency multivariate volatility model as the selected model,the Shanghai and Shenzhen 300 index and stock index futures as the empirical data,and comparing the out-of-sample prediction performance under a variety of commonly used loss functions,the paper points out that the combination of linear prediction based on genetic algorithm can obtain statistically significant prediction accuracy during the period of stable and concussion of exponential change,which is a more robust choice than using a single multivariate volatility model.

关 键 词：预测结合 多元波动率模型 遗传算法 模型置信集检验 动态套期保值 

分 类 号：F830.91[经济管理—金融学]