基于不同样本频率时变参数β的实证研究

发布时间：2018-06-12 10:02

本文选题：MTPCOV + KCOV　；参考：《西南财经大学》2014年硕士论文

【摘要】：资本资产定价模型(CAPM)是现代金融学的理论基础,这一模型在证券估价、投资组合等方面有着广泛的应用,CAPM中最具突破性意义的是β系数,它描述了单个资产受市场整体波动的影响程度,刻画了证券资产的系统性风险。由于参数β在投资理论和实践中有着重要的地位,因此对参数β的准确估计具有极其重要的理论价值和现实意义。关于参数β估计的文献中,主要存在两方面的不足：(1)文献中通常都是基于低频数据,运用高频数据估计参数β的比较少。金融市场上的信息对资产价格的影响是连续的过程,当数据采样频率比较低时,就会导致数据的离散,数据的离散会导致不同程度的信息缺失。(2)随着计算机的发展,高频数据的获得成为可能,学者们基于高频数据提出了很多高频协方差矩阵的估计方法,部分文献也将这些估计方法运用到参数β的估计上。在复杂的金融市场上,所观测到的高频资产价格往往包含市场微观结构噪声和跳跃的成分,微观结构噪声对资产价格的影响程度随着数据的采样频率的提高而加强。只考虑市场微观结构噪声和跳跃两者之一对资产协方差矩阵的影响往往是不全面的,就无法准确的估计出高频协方差矩阵。然而目前文献对参数β估计所选用的高频协方差矩阵估计方法仅仅考虑市场微观结构噪声或跳跃,没有综合考虑两个方面的影响。针对现有文献所描述的不足之处,本文基于高频数据,运用综合考虑金融市场微观结构噪声和跳跃影响的高频协方差矩阵估计方法：修正的门限预平均已实现协方差阵(MTPCOV)来估计参数β, MTPCOV是用修正的预平均方法消除微观结构噪声、用门限的方法剔除跳跃,通过这个方法估计出来的参数β理论上比目前文献中的其它估计方法要更加精确。为了验证这一点,本文分别基于低频和高频数据来对参数β进行估计,对于低频数据,选择基于状态空间模型的Kalman滤波算法和DCC-MVGARCH模型；对于高频数据,选择多元核光滑协方差阵(KCOV),运用四种方法对参数β进行估计,并且用估计出来的时变参数β建立约束条件,构建动态投资组合,通过比较投资组合在分散非系统性风险和最大化组合收益率方面的效果,来分析时变参数β约束的有效性,从而可以比较四种方法估计参数β的精确性。本文论述了文章选题的背景和意义,阐述了本文的研究思路、方法与创新,对近年来的相关文献进行了梳理,对文章的章节安排、主要研究内容进行了说明。本文选取上证180指数作为市场组合数据,时间跨度从2012年1月4日到2013年12月31日,按照时间跨度内上证180指数中180支样本股的日度交易频率高低均匀选取8支股票,以日度交易数据和1分钟高频交易数据作为研究样本。通过对时变参数β进行实证研究,本文得出： (1)参数β是不稳定的,具有时变的特性。(2)DCC-MVGARCH模型、KCOV以及MTPCOV估计出来的参数β波动比较大,运用状态空间模型估计出来的时变参数β波动较小。(3)MTPCOV估计出来的时变参数β约束下的动态市场组合的累计收益率大于KCOV、DCC-MVGARCH模型以及状态空间模型估计出来的时变参数β约束下的动态市场组合；所构建的动态市场组合的系统性风险占总风险的比例也大于其它三种方法估计出来的时变参数β约束下的动态市场组合。
[Abstract]:The capital asset pricing model (CAPM) is the theoretical basis of modern finance. This model is widely used in securities valuation, investment portfolio and so on. The most breakthroughs in CAPM is the beta coefficient. It describes the impact of the volatility of a single asset by the market as a whole and portrays the systematic risk of securities assets. Capital theory and practice play an important role. Therefore, accurate estimation of parameter beta is of great theoretical and practical significance.
There are two main deficiencies in the literature on parameter beta estimation: (1) there are usually low frequency data based on low frequency data and low estimation of parameters by high frequency data. The impact of information on asset prices in financial markets is a continuous process. When the ratio of data sampling frequency is low, the data will be dispersed and the data is discrete. (2) with the development of computers, the acquisition of high frequency data is possible, and many high frequency covariance matrices are proposed based on high frequency data. Some literature also applies these methods to the estimation of parameter beta. In complex financial markets, the high frequency assets are observed. The price often includes the components of the market micro structure noise and jumping, and the influence of the microstructural noise on the asset price is strengthened with the increase of the sampling frequency of the data. Only considering the influence of the market micro structure noise and jumping both on the asset covariance matrix is often not comprehensive, and the high frequency can not be accurately estimated. Covariance matrix. However, the high frequency covariance matrix estimation method used in the current literature on parameter beta estimation only takes into account the market microstructure noise or jump, and does not consider the effects of two aspects.
In view of the shortcomings described in the existing literature, based on the high frequency data, this paper uses the high frequency covariance matrix estimation method which considers the microstructural noise and jumping effects of the financial market synthetically: the modified threshold preaverage has realized the covariance matrix (MTPCOV) to estimate the parameter beta, and the MTPCOV is to eliminate the microstructural noise by the modified preaverage method. The method of thresholding is used to eliminate hopping. The parameter beta estimated by this method is more accurate than other estimation methods in the literature.
In order to verify this point, this paper estimates the parameter beta based on the low frequency and high frequency data respectively. For low frequency data, we choose the Kalman filtering algorithm and the DCC-MVGARCH model based on the state space model. For the high frequency data, the multiple kernel smooth covariance matrix (KCOV) is selected and the parameter beta is estimated with four methods, and the estimation is used. The time varying parameter beta builds the constraint conditions, constructs the dynamic portfolio, and compares the effectiveness of the time-varying parameter beta constraint by comparing the effect of the investment portfolio in dispersing the non systematic risk and maximizing the combination yield, thus comparing the four methods to estimate the accuracy of the parameter beta.
This paper expounds the background and significance of the topic selection, expounds the research ideas, methods and innovations of this article, combs the relevant literature in recent years, and explains the main contents of the chapters in the article. This paper selects the Shanghai 180 index as the number of the market combination, the time span from January 4, 2012 to December 31, 2013, According to the daily transaction frequency of 180 sample stocks in the 180 Index of the time span, 8 shares are evenly selected, and the daily transaction data and the 1 minute high frequency transaction data are used as the research samples.
Through empirical research on time-varying parameter beta, the paper concludes that:
(1) the parameter beta is unstable and has time-varying characteristics. (2) the DCC-MVGARCH model, KCOV and MTPCOV estimate the parameter beta fluctuation relatively large, and the time-varying parameter beta fluctuation estimated by the state space model is smaller. (3) the cumulative yield of the dynamic market combination under the time-varying parameter beta constraint under the MTPCOV estimation is greater than KCOV, DCC-MVGARCH The dynamic market combination under the constraint of time-varying parameter beta is estimated by the model and the state space model. The proportion of the systematic risk of the dynamic market portfolio to the total risk is greater than the dynamic market combination under the time variable parameter beta constraint estimated by the other three methods.
【学位授予单位】：西南财经大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F224;F830.91

【参考文献】