VaR度量的半参数方法研究.pdf

摘 要 本文主要提出了一种估计在险价值（VaR ）的新方法—半参数 VaR 度量法， 新的半参法同时汲取了参数法对波动率的捕捉以及非参法避免模型假设错误的 优点，可以提高 VaR 估计的准确度。 VaR 风险度量从诞生便受到广泛关注，其估计方法也层出不穷。大致分为 参数法--需要事先对目标序列进行某些统计假定，代表方法是方差-协方差法、 GARCH 类模型估计等；以及非参数法--不需要事先假定分布，代表方法是历史 模拟法、蒙特卡罗模拟法。这两类方法都各有优缺点，也出现了很多改进的措 施。 本文提出的半参数 VaR 度量是参数法与非参数法的结合。参数法估计是利 用 GARCH 类模型估计目标序列的波动性。本文借助贝叶斯推断方法，利用 MCMC 模拟引入了 GARCH 模型不同于传统 MLE 的估计方法—BayesGARCH 估计，提高了模型估计精度。同时还在参数估计的 MCMC 迭代过程中多次对目 标序列分位点估计，从而形成目标序列分位点的非参数估计。综合对序列波动 性的参数估计和分位点的非参数估计得出 VaR 值。 最后，本文以伦敦 Brent 原油期货近四年价格为基础，将本文提出的 VaR 度量方法同经典的度量方法进行了比较，证实了本模型的优越性。 关键词：在险价值（VaR）；半参法；GARCH；贝叶斯推断；马尔科夫蒙特卡罗模 拟（MCMC） Abstract This paper proposes a new method to estimate value at risk(VaR), semi-parameter VaR. Because of combining the parameter methods advantages of capturing volatility and the non-parameter methods merits of advoiding false model assumption, this method can improve the accuracy the VaR estimation method. VaR has been received extensive attention as the day it was born. Its estimation methods are endless. On the whole, it can be divided into two types, that is parameter methods and non-parameter methods. Parameter methods need to make some statistical assumptions for the target series in advance, and their representative methods include variance-covariance method and the class of GARCH models method. Non-parameter methods dont need to assume the distribution first, and their representative methods include the historical simulation method and Monte Carlo simulation. This two kinds of methods both have their advantages and disadvantages, and many improvements measures have been appearing. The semi-parameter VaR measure combines the parameter method and non-parameter metho