基于APARCH和POT模型的上证综指风险度量

发布时间：2018-11-04 16:46

【摘要】：近年来金融市场迅猛发展的同时也伴随着波动性增强,并致使许多金融危机事件频繁发生。金融监管机构和众多的投资者因此也增加了对市场风险的关注度。如何寻找到一种更精准的度量工具成为了当前度量市场风险的首要问题,风险价值(VaR)作为目前最普遍的风险度量工具,它表示在一定的置信水平下,资产或资产组合在未来某一特定时间内的最大可能损失,也即是给定显著性水平资产组合收益损失分布函数的分位数点。传统VaR中假定收益率服从某一特定分布,鉴于此极值理论中POT模型逐渐成为VaR估计的主流方式之一。这是因为POT模型只需要研究收益率序列的尾部特征,利用GPD分布来拟合尾部分布。本文在POT模型基础上引入APARCH模型,将二者结合研究上证市场风险,这在当前主流观点认为关注极端VaR的尾部风险比关注正常情况下的风险更有实际意义的背景下,本文对如何有效准确地度量市场风险进行了有益探索。本文以上证指数1990年12月19日-2012年3月19日日收盘价为原始数据,在借鉴吸收前人成功的研究成果同时,采用实证分析方法对上海证股票市场风险进行分析,并比较研究了不同置信水平下VaR估计值。首先,本文对收益率序列的统计性特征进行了描述,以便为选择合理的VaR估计模型。通过正态性、自相关性以及ARCH效应的检验,本文发现我国股市收益率序列具有尖峰厚尾性,弱自相关性,波动集聚性。其次,本文采用APARCH模型捕捉收益率序列的自相关和异方差现象,并采用极大似然法估计模型参数,因为极大似然估计需假设残差的分布,故而使用GMM估计对MLE估计结果进行校正。最终获得近似独立同分布的残差序列。再利用POT模型对经过ARARCH模型筛选过的残差进行极值分析,并根据VaR的可加性计算出收益率序列在不同置信水平下的VaR。为了比较研究,本文使用一般POT模型估计了上证综指的VaR,由于一般POT模型不能避免超出量序列的相关性,故而可能会高估真实的市场风险。通过比较不同置信水平下两模型VaR的值,可以得出以下结论： 1.在不同的显著性水平下,一般的POT模型估计出的均VaR高于APARCH-POT模型估计出的VaR值,这表明一般的POT模型的确高估了市场风险,且经APARCH—POT模型的估计结果更保守,这也大大了提高尾部分布VaR的稳定性。 2.本文通过使用Kupiec失败返回检验法对各VaR值进行了有效性检验并发现：POT模型和经APARCH过滤后的POT模型在95%、99%的置信水平上Kupiec检验结果均是有效的。但在99%置信水平上POT模型检验效果更好。 3.通过对使用APARCH过滤后的POT模型进行动态建模得到了整个样本期内的VaR值。根据样本期间的VaR分布可以得出以下结论：样本期的VaR按集聚特征分为五个区间,其中1990-1994年,沪市建立之初,政府对股市的态度决定了此时市场风险,风险值最大；1995-1999年,随着政府对发展股市信心的坚定,市场风险开始回落,但整体风险仍较大；2000-2006年,随着制度的成熟和机构投资者的进入,市场风险进入平缓期,风险值相对较低；2006-2010年频繁的经济刺激政策推高市场风险；2010-至今,政策打压下投机泡沫减少,上个区间段风险再次回落。本文的主要创新点在于使用APARCH模型对过滤收益率序列建模时,对比分析了扰动项服从不同分布时APARCH模型捕捉收益率特征的有效性。另外,本文采用APARCH模型和POT模型相结合的方式拟合收益率,避免了传统POT模型由于数据相关而造成VaR高估的问题。
[Abstract]:With the rapid development of financial markets in recent years, the volatility has been accompanied by the frequent occurrence of many financial crisis events. Financial regulators and many investors have therefore increased attention to market risks. How to find a more precise measurement tool has become the primary problem of the current measure market risk, and the risk value (VaR) is the most common risk measure tool at present, which means that at a certain confidence level, The maximum possible loss of a portfolio of assets or assets over a certain period of time, i.e., a fractional number of digits of the distribution function for the combined benefit of a given significance level asset portfolio. In the traditional VaR, we assume that the yield is subject to a certain distribution. In view of this extreme value theory, the POT model gradually becomes the mainstream of VaR estimation. This is because the POT model only needs to study the tail feature of the yield sequence and use the GPD distribution to fit the tail. In this paper, we introduce the APARCH model on the basis of the POT model, and combine them in the research of the market risk, which in the current mainstream view is concerned that the tail risk of extreme VaR is more meaningful than the risk of paying attention to the normal situation. Under the scene, this article is helpful for how to measure market risk accurately and accurately In this paper, from December 19, 1990 to March 19, 2012 as raw data, the paper analyses the risk of Shanghai Shanghai Stock Market by using the empirical analysis method, and compares the V at different confidence levels. First of all, this paper describes the statistical characteristics of the yield sequence, so as to select reasonable price. By means of positive state, self-correlation and ARCH effect, this paper finds that the stock market rate of return on stock market has peak-thickness tail, weak self-phase, Secondly, we use the APARCH model to capture the autocorrelation and heteroscvariance of the yield sequence, and estimate the model parameters by using the maximum likelihood method, because the distribution of the residual is assumed to be assumed greatly, thus the GMM estimation is used. The MLE estimation result is corrected. Finally, near We use POT model to analyze the residual error filtered through ARARCH model, and calculate the yield sequence according to the additivity of VaR. In order to compare the VaR under different confidence levels, the VaR of Shanghai Composite Index is estimated using the general POT model. So it's possible to overestimate the real market risk. By comparing two models at different levels of confidence Based on VaR, the following conclusions can be obtained: 1. Under different saliency levels, the average VaR of the general POT model is higher than the VaR value estimated by the APARCH-POT model, which indicates that The general POT model does overestimate the market risk, and the estimation results of the APARCH-POT model are more conservative, In this paper, the validity of VaR is verified by using Krupiec failure return test. The POT model and POT model after APARCH filtering are 95% and 99% respectively. The Krupiec test results are valid at the confidence level. The POT model test effect is better at 99% confidence level. 3. By using the PO filtered by APARCH The VaR of the whole sample period is obtained by the dynamic modeling of the T model. Based on the VaR distribution during the sample period, the following conclusions can be obtained: the VaR of the sample period is divided into five sections according to the accumulation characteristics. In 1990-1994, the attitude of the government to the stock market determines the market risk at this time, and the risk value is the largest; 19 From 95 to 1999, as the government is firm in developing stock market confidence, the market risk begins to fall back, but the overall risk is still large; in 2000-2006, with the maturity of the system and the entry of institutional investors, the market risk is relatively low, and the risk value is relatively low; 2006-2010 Frequent economic stimulus policies pushing high market risk; 2010-present, The main innovation point in this paper lies in the comparison and analysis of the disturbance items when the APARCH model is used to model the filtering rate of return. In addition, we use APARCH model and POT model to fit the yield and avoid
【学位授予单位】：东北财经大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：F224;F832.51

【参考文献】