Abstract: Compared with low-frequency volatility models, high-frequency volatility models can achieve moreaccurate volatility and risk forecasts of single assets, thus how to introduce high-frequency volatility models into portfolio risk analysis has a theoretical and practical significance. Taking the high frequency data of six various industries in CSI 300 index as an example, we construct nine HAR-RV-type models to depict industry index volatility using the out-of-sample rolling time window technique. At the same time, R-vine model is built to describe the dependence structure between industry indexes. On that basis, combining the optimized weight of industry assets from Mean-CVaR method, the expected shortfall models of industry portfolio are constructed and the backtesting method is used to compare the accuracy of different risk models. The results prove that the HAR-type models which are the crucial proxies of high-frequency volatility models can effectively forecast the expected shortfall of industry portfolio. Then, the high-frequency volatility forecasts can further affect the accuracy of portfolio risk forecasts. The jump, signed jump variation and signed negative and positive jump variation are helpful to forecast industry portfolio risk more accurately.

Key words: industry portfolio, expected shortfall, R-vine, realized volatility, HAR-type models, backtesting

摘要： 相较于低频波动率模型,高频波动率模型在单资产的波动和风险预测中均取得了更好效果,因此如何将高频波动率模型引入组合风险分析具有重要的理论和现实意义。本文以沪深300指数中的6种行业高频数据为例,运用滚动时间窗技术建立9类已实现波动率异质自回归(HAR-RV-type)模型刻画行业指数波动,同时使用R-vine copula模型描述行业资产间相依结构,进一步结合均值-CVaR模型优化行业资产组合投资比例,构建组合风险的预期损失模型,并通过返回测试比较不同风险模型的精度差异。研究结果表明:将HAR族高频波动率模型引入组合风险分析框架,能够有效预测行业资产组合风险状况;高频波动率预测的准确性将进而影响组合风险测度效果,跳跃、符号跳跃变差以及符号正向、负向跳跃变差均有助于提高行业组合风险的预测精度。

关键词: 行业组合, 预期损失, R-vine, 已实现波动, HAR族模型, 返回测试