基于流动性调整的高频协方差阵的估计及其应用研究  

Estimation and Application Study on Covariance Matrix of High Frequency Data Based on Liquidity Adjustment

作　　者：刘丽萍[1] 马丹[2] 

机构地区：[1]贵州财经大学数学与统计学院,贵州贵阳550025 [2]西南财经大学统计学院,四川成都610071

出　　处：《管理工程学报》2016年第2期76-83,共8页Journal of Industrial Engineering and Engineering Management

基　　金：国家社会科学基金资助项目(10XTJ0001);贵州财经大学人才引进资助项目

摘　　要：金融资产交易往往不具有时间的一致性,采用高频数据估计协方差阵时需要避免由于异步交易导致的"Epps"效应。常用的时间刷新技术能够解决异步交易问题,但随着资产数量增加,样本量会迅速减少。本文介绍了基于流动性调整的双频协方差阵估计方法(Rn BTSCOV),该方法可减少数据量的损失,在不对参数施加任何限制的情况下,提高估计精度。将该方法应用到投资组合中与常用的已实现协方差阵和双频协方差阵进行对比分析,研究发现Rn BTSCOV方法在所有的标准下具有更好的表现。The covariance matrix of financial assets plays an important role in portfolio management. Because high-frequency data contains richer information, an increasing number of scholars began to consider using high-frequency data to estimate the covariance matrix of financial assets. However, the trading time of financial assets is usually inconsistent. When using high-frequency data to estimate the covariance matrix, we need to avoid ＂Epps＂ effect which is caused by asynchronous transactions. The refresh time technology is commonly used to solve asynchronous transaction problems. However, with the increasing number of assets the sample size will decrease rapidly. To reduce the amount of data loss and improve the estimation efficiency of covariance matrix, we use blocking strategy and regularization approach to estimate TSCOV, and propose Rn BTSCOV estimator which is based on liquidity adjustment. The blocking strategy starts by ordering assets in the covariance matrix according to transaction frequency, with the most liquid assets in the top left corner and the least liquid assets in the bottom right corner. This initial step ensures that subsequent blocks will group assets with similar transaction frequencies. We further divide assets into liquidity-based clusters. Asset clusters are then combined to form a BTSCOV at last, where each block itself is a covariance matrix. Because TSCOV is not necessarily positive definiteness, the BTSCOV, which is reconstructed by a series of small TSCOV, is also not necessarily positive definiteness. Positive definiteness of the covariance matrix is a very important theoretical nature. When high frequency covariance matrix applies in the portfolio and risk management and if the covariance matrix does not have positive definiteness, combinatorial optimization problem becomes very difficult. In order to ensure positive definiteness of the covariance matrix, we adopted the ＂Eigenvalue Cleaning＂ technique and then obtained the ＂Rn BTSCOV＂. We apply the Rn BTSCOV estimator in

关 键 词：双频协方差阵 基于流动性调整的双频协方差阵 等比例风险投资组合 

分 类 号：F830.9[经济管理—金融学]