几种不同类型金融时间序列模型预测的比较研究

发布时间：2018-03-09 04:35

本文选题：中位数估计　切入点：AR(1)模型　出处：《中国矿业大学》2017年硕士论文　论文类型：学位论文

【摘要】：时间序列分析是统计学中一个重要分支,近些年来各种时间序列模型在各个领域被广泛用于数据分析预测,特别在金融领域.AR模型是最早的金融时间序列模型,对AR模型自回归参数估计方法很多,估计的效果也不尽相同.而GARCH模型可以说是为金融领域量身定做的时间序列模型,不同分布下的GARCH类模型对股市刻画及风险价值预测中效果和特点不尽相同.因此,对AR模型和GARCH类模型预测研究具有重要意义.本文主要研究改进后的AR(1)模型预测区间在模拟中的效果及对AR(1)模型参数估计方法及预测研究,同时利用不同GARCH模型对深圳综指和上证指数进行实证研究.首先,对AR(1)模型自回归参数采用中位数估计构造该模型预测区间,一种是不带单位根检验的,另一种带单位根检验.然后进行蒙特卡罗模拟实验,通过比较发现样带单位根检验的预测区间无论从区间覆盖率还是平均长度方面效果都好一些.接着在对AR(1)模型采用不同参数估计方法进行建模分析,进而采用不同预测方法进行预测,发现在平稳状态下最小二乘估计估计效果良好.其次,基于不同GARCH模型下的深市收益率研究表明该股市具有杠杆效应,通过比较发现EGARCH-t模型能够很好刻画深市收益率序列特征.同时采用不同分布下的IGARCH和TGARCH模型对上证指数进行研究,发现GED分布下的TGARCH模型能够很好的刻画上证收益率序列特征,进而计算出各个模型下的风险价值,接着对计算出来的VaR进行了Kupiec检验,通过检验可以看出GED分布下的TGARCH模型在风险度量方面效果最好.最后,讨论一下目前存在的一些相关问题及未来研究的展望.
[Abstract]:Time series analysis is an important branch of statistics. In recent years, various time series models have been widely used in data analysis and prediction, especially in the financial field. There are many methods to estimate the autoregressive parameters of AR model, and the effect of the estimation is not the same. The GARCH model can be said to be a time series model tailor-made for the financial field. GARCH models with different distributions have different effects and characteristics on the description of stock market and the prediction of risk value. It is of great significance to study AR model and GARCH model prediction. In this paper, we mainly study the effect of the prediction interval of improved ARH-1 model in simulation, and the estimation method and prediction method of ARQ1 model parameters. At the same time, different GARCH models are used to study the Shenzhen Composite Index and Shanghai Stock Exchange Index. Firstly, the median estimation is used to construct the prediction interval of ARQ1 model, one is without unit root test. Another test with unit roots. Then Monte Carlo simulation, It is found that the prediction interval of sample band unit root test is better in terms of interval coverage and average length. Then different parameter estimation methods are used to model and analyze ARG-1 model. Then different forecasting methods are used to predict, and it is found that the estimation effect of least square estimation is good in stationary state. Secondly, the deep market yield research based on different GARCH models shows that the stock market has leverage effect. It is found that the EGARCH-t model can well describe the characteristics of the yield series of Shenzhen Stock Exchange. At the same time, the IGARCH and TGARCH models under different distributions are used to study the Shanghai Stock Exchange Index. It is found that the TGARCH model under the GED distribution can well describe the characteristics of the yield series of the Shanghai Stock Exchange. Then the risk value of each model is calculated, and then the calculated VaR is tested by Kupiec. It can be seen that the TGARCH model under the GED distribution has the best effect on risk measurement. Finally, This paper discusses some related problems and prospects for future research.
【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】：F224;F832.51

【参考文献】