基于改进GARCH-MIDAS模型的宏观经济因素影响股价波动研究

发布时间：2018-04-16 11:13

本文选题：股市波动 + GARCH-MIDAS　；参考：《中国矿业大学》2017年硕士论文

【摘要】：目前在经典计量经济学研究方法中主要体现两种现象:一方面在研究过程中均使用相同频率的数据;另一方面许多研究以低频率的股票市场数据为研究对象的数据指标,使得我国股票市场数据与宏观外生解释变量的数据具有相同的频率。在对不同频率的时间序列数据进行处理时,通常将高频率的数据转化为低频率或者同频率的数据,这极有可能损失混频(高频)数据包含的信息有效性。本文正是在这样的情况下提出以混频数据作为研究对象的数据指标,既包含低频数据又包含高频数据。传统的同频模型由于受到数据频率的限制,不能得到宏观经济解释变量和我国股票市场波动之间的因果关系,Engle提出的经典广义自回归条件异方差混频数据抽样模型称作GARCH-MIDAS(Generalized AutoRegressive Conditional Heteroskedasticity),将混合抽样技术方法(Mixed Data Sampling)融入传统的GARCH模型中,将波动率分解为短期和长期因素,并巧妙地将宏观经济因素作为长期波动率的解释因子,使数据能够得到更充分地利用。而本文与经典GARCH-MIDAS模型有所不同,本文在解释变量加入美元兑人民币汇率高频数据,并采用美元兑人民币汇率的已实现波动率来解释我国股市长期波动,拓展经典GARCH-MIDAS模型。在实证研究中,分别从水平效应和波动效应两个角度建立多个单因素和多因素GARCH-MIDAS模型对我国股市波动进行分析估计。本文主要选取月度货币供应、月度消费者价格指数和日度汇率为指标研究股市波动。基于改进GARCH-MIDAS模型的研究结果显示:货币供应量的水平值和波动率均对我国股票市场波动有显著的正向影响关系。消费者价格指数的水平效应与我国股票市场波动有显著的负相关关系,而在波动效应上与我国股票市场波动之间的关系不显著。美元兑人民币汇率在水平效应和波动效应均对股票市场波动产生显著的负相关关系。多因素模型的估计结果与单因素模型的估计结果基本相同,但是由于多因素模型估计的参数较多,可能存在过度参数化等问题,这些问题使得一些系数不再显著。同时,对研究模型进行预测能力分析发现单因素模型和多因素模型均有很强的预测能力,而且基于多因素水平效应模型的预测效果比基于单因素水平效应模型的预测效果更好。此外,通过比较发现多因素混频模型比单因素混频模型可以更好地刻画我国股票市场价格波动的长期成分。结合我国实际情况提出以下政策建议,包括提高宏观调控政策效力、发展健康股市、引导理性投资及完善披露机制等。
[Abstract]:At present, there are two main phenomena in classical econometrics research methods: on the one hand, the data of the same frequency are used in the research process; on the other hand, many studies take the low-frequency stock market data as the data index.It makes the data of Chinese stock market have the same frequency as the data of macroscopical exogenous explanatory variables.When processing time series data with different frequencies, the high frequency data is usually converted to low frequency or the same frequency data, which is likely to lose the validity of the information contained in the mixing (high frequency) data.In this paper, the data index of mixing data is proposed, which includes both low frequency data and high frequency data.The traditional co-frequency model is limited by the data frequency.The causality between macroeconomic explanatory variables and stock market volatility can not be obtained. The classical generalized autoregressive conditional heteroscedasticity mixed data sampling model proposed by Engle is called GARCH-MIDAS(Generalized AutoRegressive Conditional heteroscedastic sampling. The mixed Data sampling method is described as mixed Data sampling.Into the traditional GARCH model,The volatility is decomposed into short and long term factors, and macroeconomic factors are used as the explanation factors of long term volatility, so that the data can be used more fully.This paper is different from the classical GARCH-MIDAS model in explaining the variables by adding the high frequency data of USD / RMB exchange rate and using the realized volatility rate of USD- RMB exchange rate to explain the long-term volatility of China's stock market and extend the classical GARCH-MIDAS model.In the empirical study, we establish several single-factor and multi-factor GARCH-MIDAS models from the perspective of horizontal effect and volatility effect to analyze and estimate the volatility of China's stock market.This paper mainly selects monthly money supply, monthly consumer price index and daily exchange rate as indicators to study stock market volatility.The results based on the improved GARCH-MIDAS model show that both the level of money supply and the volatility have significant positive effects on the volatility of China's stock market.There is a significant negative correlation between the horizontal effect of the consumer price index and the volatility of the stock market in China, but there is no significant relationship between the fluctuation effect and the volatility of the stock market in China.Dollar / RMB exchange rate has a significant negative correlation with stock market volatility both in horizontal effect and volatility effect.The estimation results of multivariate model are basically the same as those of single factor model. However, due to the large number of parameters estimated by the multivariate model, there may be some problems such as over-parameterization, which make some coefficients less significant.At the same time, it is found that both the single-factor model and the multi-factor model have strong predictive ability, and the prediction effect based on the multi-factor horizontal effect model is better than that based on the single-factor horizontal effect model.In addition, it is found that the multi-factor mixing model can better describe the long-term components of the stock market price volatility in China than the single-factor mixing model.According to the actual situation of our country, this paper puts forward the following policy suggestions, including improving the effect of macro-control policy, developing healthy stock market, guiding rational investment and perfecting the disclosure mechanism, etc.
【学位授予单位】：中国矿业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.51;F124;F224

【参考文献】