基于MIDAS-QR的均值-VaR参数化组合投资决策

Mean-VaR Parametric Portfolio Selection via the MIDAS-QR Method

文章将混频数据分析和特征变量参数化的思想引入组合投资决策,构建基于MIDAS-QR的M-VaR参数化组合投资模型与方法,克服了传统均值-VaR (M-VaR)模型忽略动态市场环境、混频数据信息和金融资产特征等方面的不足.文章的核心技术和主要创新在于:第一,利用频率对齐技术和多项式权重约束函数,将高频资产特征引入组合投资权重函数;第二,引入参数化组合投资策略,考虑金融资产特征变量并将其参数化,在M-VaR框架下构建参数化组合投资模型;第三,通过理论推导,将M-VaR参数化组合投资转化为MIDAS-QR回归问题,给出模型求解方法.选取中国股市行业板块指数进行实证研究,结果表明:文章在混频数据环境下构造的参数化组合投资模型,建立了组合投资权重与高频金融资产特征之间联系,增加了组合投资结果的可解释性;新模型能够充分挖掘高频特征变量信息,给出时变的组合投资权重,实现动态组合投资选择,获得了比其他模型更好的组合投资绩效;基于MIDAS-QR的求解方法,有效缩减了模型中待估计参数数目,能够解决大规模动态组合投资决策问题.

This paper introduces mixed-frequency data analysis and the notion of parameterization of features into portfolio optimization,and develops a mean-VaR parametric portfolio selection model using the MIDAS-QR method.This novel model overcomes the deficiencies of conventional mean-VaR(M-VaR) model which ignores the dynamic market environment,mixed-frequency data information and asset features.The core technologies and main innovations of this paper are as follows.First,the original high-frequency characteristic variables are introduced into the portfolio weight function using the frequency alignment technique and polynomial weight constraint function.Second,under the mean-VaR framework,introducing parametric portfolio policies,the parametric portfolio selection model is constructed with characteristic variables.Third,the mean-VaR parametric portfolio selection model is transformed to an MIDAS-QR question theoretically.For an illustration purpose,an empirical study is conducted on the industry sector index of Chinese stock market.The empirical results show that the novel model provides a link between portfolio weights and characteristic variables,which improves the interpretability of the resulting portfolios.It also fully captures the useful information contained in high-frequency characteristic variables,provides time-varying portfolio weights for dynamic portfolio selection,and obtains better portfolio performance than other models.Moreover,the solution method based on MIDAS-QR effectively reduces the number of parameters to be estimated,is efficient for solving large-scale dynamic portfolio selection.