基于GARCH模型VAR方法外汇风险度量

发布时间：2018-01-21 23:25

本文关键词： 外汇风险 GARCH类模型 VAR方法偏T分布　出处：《山东大学》2013年硕士论文　论文类型：学位论文

【摘要】：自2005年7月21日中国实施汇率体制改革之后,人民币汇率的变动越来越频繁；同时,随着中国加入WTO之后,在国际货币市场、资本市场所占有的份额也越来越大；中国已成为第一大外汇储备国,但迫于国际政治势力和经济因素,人民币正面临着升值的压力,这样一来对我国的外汇储备将会在国际汇率市场上迅速贬值。在这样的环境下,对外汇风险的管理就显得尤为重要。 VAR方法是风险管理中主流风险度量方法,已被各国尤其是在西方发达国家的金融机构和企业运用。在中国,该方法大都运用在相对比较成熟的股票市场,真正运用在外汇市场的情况还不多。因此,本文尝试着根据中国外汇市场收益率的统计特征,运用基于各类GARCH类模型的VAR方法寻找一种适合我国具体情况的外汇风险度量方法。本文通过对美元收益率R序列的特征分析发现序列存在着非正态性、尖峰厚尾、非对称性、非独立性、波动集束、条件方差时变性以及长记忆性等特征。使用传统的时间序列模型ARMA模型对序列拟合,经过检验ARMA模型不能处理序列的这些特征。由GARCH类模型的性质利用GARCH类模型对R序列拟合,各种GARCH类模型均能很好的处理收益率序列的异方差效应。为了综合考虑各类模型对收益和损失的预测情况,分别计算高位95%、99%和低位5%、1%、0.5%以及0.25%基于各种GARCH类模型的VAR值,并进行准确性检验：在低位随着置信水平的增加,基于正态分布的各类GARCH模型都对VAR值高估(低估风险)越来越严重,说明了相对于R序列正态分布的左侧尾部太薄；在高位95%、99%,基于T以及GED分布各类GARCH类模型的VAR值都出现了高估(高估收益),即相对于R序列分布的右侧尾部较厚,但基于偏T分布的各类GARCH模型则基本都通过了检验,但对于下侧VAR值,基于T分布的GARCH类模型对VAR值随着置信水平的增加有一定的高估(低估风险),说明T分布的左侧尾部相对于R序列稍薄；GED分布的GARCH类模型对VAR值有一定的低估(高估风险),随着置信水平的增加,高估程度越来越大,说明GED分布的左侧尾部相对于R序列稍厚,说明R序列确实存在左偏的性质,左侧尾部较厚,这也与近年了美元相比于人民币呈现贬值的趋势是一致的。但是基于偏态T分布的GARCH类模型对下侧VAR值的预测随着置信水平的增加,VAR值的低估(高估风险)越来越严重；虽然偏T分布对于上侧VAR值的预测与期望值较为接近,但在风险管理过程,风险比收益重要,所以在最优模型的选择上,还是以下侧VAR值的预测结果为标准,基于T分布和GED分布的各模型对于下侧VAR值的预测较为稳定,均值方程中带有条件标准差模型优于均值方程中不带有条件标准差的模型,说明在外汇市场中收益是还是和风险紧密相连的,考虑对下侧VAR值的预测值与期望值的接近程度,基于T分布的GARCH-M模型对VAR值的预测最接近期望值,且随着置信水平的增加相对来说比较稳定。为了与GARCH类模型作比较,计算了基于ARMA模型的下侧VAR值,并对其做准确性检验。从结果可以看出基于T分布的GARCH-M模型的对VAR值的估计的准确性要远远优于基于ARMA模型计算出的VAR值。
[Abstract]:After the implementation of the exchange rate system reform since July 21, 2005 China, changes in the RMB exchange rate is more and more frequent; at the same time, with the China after joining WTO, in the international monetary market, capital market share is growing; China has become the country's largest foreign exchange reserves, but under the international political and economic factors, the RMB appreciation is facing the pressure, as a result of China's foreign exchange reserves will depreciate rapidly in the international exchange market. Under such circumstances, the exchange risk management is particularly important.
The VAR method is the mainstream risk risk management measures, is in the world especially in the western developed countries, financial institutions and enterprises to use. In this method China, mostly used in relatively mature stock market, the real application in the foreign exchange market situation is not enough. Therefore, this paper try according to statistical characteristics of market returns Chinese the foreign exchange rate, using the VAR method of GARCH model based on finding a suitable to the situation of China's foreign exchange risk measurement method.
In this paper, through analyzing the characteristics of R sequence in the US yield found sequences exist non normality, leptokurtic, non symmetry, non independence, volatility cluster, time-varying conditional variance and long memory characteristic. Using the traditional time series ARMA model of sequence fitting, through the characteristic test of ARMA model can not handle the sequence. By the nature of the GARCH model with GARCH model to fit the R sequence, GARCH model can also heteroscedasticity effect processing good income rate series. In order to consider the various models of profit and loss forecast, calculate high of 95%, 99% and 5% lower, 1%. 0.5% and 0.25% kinds of GARCH model based on the VAR value, and the accuracy of the test: at a low level with the confidence level increases, the normal distribution of all kinds of GARCH models are based on the VAR value of overvalued (underestimate risk) is more and more serious, said Clear R sequences relative to normal distribution of the left tail is too thin; high in 95%, 99%, T and GED based on the distribution of all kinds of GARCH model VAR value appeared overvalued (overestimateearnings), namely R with respect to the right side of the rear sequence distribution is thick, but all kinds of partial GARCH model based on T distribution is basically through the test, but for the lower VAR value, the GARCH model of T distribution with the confidence level increases with some overestimation of VAR value based on T (underestimate risks), the distribution of the left tail relative to the R sequence of GARCH model is slightly thin; the distribution of GED have a low value of VAR (overestimate risk). With the increase of the confidence level, increasinglyovervalued. GED, the distribution of the left tail relative to the R sequence slightly thick, R series have left the nature of the left tail is thick, it is also associated with recent dollar compared to the RMB devaluation trend is presented The same. But the prediction model of GARCH skewed T distribution on the lower side based on the VAR value with the confidence level increases, the value of VAR (underestimate overestimate the risk is more and more serious); although the skewed T distribution for the prediction and expectation value is close to the VAR side, but in the process of risk management, risk return than important. So in the optimal model selection, or to predict the following side VAR value as the standard, T distribution and GED distribution of the model for predicting lower VAR value is more stable based on the mean equation with conditional standard deviation conditions standard deviation model with no model is better than the mean equation, illustrate the gains in the foreign exchange market is or risk closely linked, consider the forecast to the lower VAR value and expected value close to the degree, the GARCH-M model of T distribution prediction of VAR value of the expectations based on, and with the increase of relative confidence level It's more stable.
In order to compare with the GARCH class model, the lower side VAR value based on the ARMA model is calculated, and its accuracy is tested. From the result, we can see that the accuracy of the VAR estimation based on the T distribution model is much better than the VAR value calculated based on the ARMA model.

【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：F832.6;F224

【参考文献】