基于VPIN模型的高频波动率预测研究

发布时间：2018-03-17 17:19

本文选题：高频　切入点：波动率预测　出处：《复旦大学》2014年硕士论文　论文类型：学位论文

【摘要】：伴随技术进步,高频金融应运而生。高频交易逐渐增加的同时,面临着低频模型失效、高频数据噪音、交易时效性等问题。高频波动率的表示、分析和预测是解决上述问题的关键,也是研究的着眼点。本文是基于高频波动率代理变量、相关预测模型的比较研究。本文采用沪深300股指期货的逐笔数据,以5分钟已实现波动率为代理变量,利用多种损失函数比较了不同预测方法的预测效率。分析过程引入VPIN作为外部信息,以降低微观结构噪音的影响；同时使用HAR-VPIN模型检验了VPIN的预测效度,解决了VPIN原有检验手段不足、预测效率受质疑的问题。本文基于分析结果,结合高频交易的风险管理实践,提出高频波动率预测模型的应用场景,分析高频波动率的应用成果。研究的主要结论是：1 VPIN的计算稳健性在高频预测研究中可控。通过不同篮子数量、起始点,以及买卖方向打标算法等参数分析VPIN对参数的敏感性,发现质疑研究的主要错误在于打标算法和时间框架的应用错误。HAR-VPIN回归模型表明,VPIN能够解释成交量信息,是波动率的主要驱动因子。2利用高频数据计算已实现波动率是较好的方法。比较几种高频波动率代理变量,建议使用5分钟作为波动率预测的样本区间,既避免了更高频率的微观结构噪音,又避免了更低频率的信息时效性损失。3HAR-VPIN模型在绝大多数损失函数下预测能力较强。比较各种预测模型发现,HAR-VPIN模型由于包括了其他模型所不具备的“外部信息”,预测误差相对较低,仅在部分损失函数度量下弱于IGARCH模型。
[Abstract]:With the development of technology, high frequency finance emerges as the times require. At the same time, the high frequency trading is gradually increasing, and it faces the problems of low frequency model failure, high frequency data noise, transaction timeliness and so on. Analysis and prediction are the key to solve the above problems, and are also the focus of the research. This paper is based on high-frequency volatility proxy variables, the comparative study of relevant forecasting models. This paper adopts the data of Shanghai and Shenzhen 300 stock index futures. The prediction efficiency of different prediction methods is compared by using a variety of loss functions using 5 minutes' realized volatility as a proxy variable. VPIN is introduced as the external information in the analysis process to reduce the influence of microstructural noise. At the same time, the HAR-VPIN model is used to test the prediction validity of VPIN, which solves the problem that the original test means of VPIN are insufficient and the forecasting efficiency is questioned. Based on the analysis results, this paper combines the risk management practice of high frequency trading. The application scenario of high frequency volatility prediction model is put forward, and the application results of high frequency volatility are analyzed. The main conclusion of the study is that the calculation robustness of 1: 1 VPIN is controllable in the high frequency prediction research. After analyzing the sensitivity of VPIN to parameters, we find that the main error of the research is the error of marking algorithm and time frame. HAR-VPIN regression model shows that VPIN can interpret the information of trading volume. It is a good method to calculate realized volatility by using high frequency data. Comparing several kinds of proxy variables of high frequency volatility, it is suggested to use 5 minutes as sample interval for volatility prediction. To avoid the higher frequency of microstructural noise, It also avoids the loss of information timeliness of lower frequency. 3HAR-VPIN model has strong prediction ability under most loss functions. Comparing various prediction models, it is found that the HAR-VPIN model includes "external information" that other models do not have. The measurement error is relatively low, It is weaker than IGARCH model only in partial loss function metric.
【学位授予单位】：复旦大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F832.51

【参考文献】