节点文献
序约束下ARCH模型最小二乘估计
The Least Square Estimate of ARCH Model under Ordered Restriction
【作者】 王晓光;
【导师】 宋立新;
【作者基本信息】 吉林大学 , 概率论与数理统计, 2004, 硕士
【摘要】 自回归条件异方差模型(Autoregressive Conditional Heteroskedasticity或ARCH)作为一个重要的非线性时间序列模型,近年来得到了迅速发展。实际上,股票价格序列的波动是金融市场中讨论得很多的主题,异方差是客观存在的现象,为了准确地刻划方差波动的这一性质,Engle在1982年开创性地提出了自回归条件异方差模型,Engle和Kraft在1983年把此模型一般化。他们假设预测误差εt为某实值离散时间随机过程,并且是某随机过程的随机扰动,即yt=g(Xt,b)+εt,其中Xt是外生变量和yt的时滞组成的向量,b是均值参数向量。记ft为截止时刻t的所有信息的信息集合,进一步假设εt是某线性回归方程的随机扰动,那么可以建立如下时间序列模型。 时间序列{Xt},满足其中α0>0,ai≥0,i=1,2,…,q,{ξt}(?)N(0,1),t=1,2,…,且ξt与{Xs,s<t}相互独立。此模型常记为ARCH(p,q)模型。 本文主要研究平稳遍历的ARCH(0,q)模型,即时间序列{Xt}满足其中α=(α0,α1,…,αq)T∈θ0={α0>0,αi≥0,i=1,2,…,q,α1+…+αq<1},Eξt=0,Eξt2=1,Eξt4<+∞,t=1,2,…,且ξt与{Xs,s<t}相互独立。吉林大学硕士学位论文序约束下ARCH模型最小二乘估计2设X卜。,瓜一。,…,X一1,X0,Xl,瓜,…n+q的一段随机样本,记信息集入一:=a我们记,抵为来自此平稳遍历的序列容量为(Xs,s(亡一1).h:(x卜1;a)一a。+al:熟1+…+a。:熟。 =a。+a1Xt--1+一+aoXI--。,L。(a)=Ln(a。,al,…,a。)=艺“(Xt一“a),(1 .2)其中‘(凡一1;a)一[对一、:(凡一1;a)]2(1 .3) 那么就可以研究使L。(a)在e。上取最小值的最小二乘估计的性质. 令。为参数。的最小二乘估计(LsE)。。的具体形式,可以通过由哭乎2}。一。- 0所得到的一组线性方程组解出.在求解d的时候,我们是通过最优化方法中的 求带约束的极值问题的Kuhn-Tucker方法得到.由于我们求的是最小二乘估计, 因此这里涉及到的最优化方法实际上是有约束的二次规划问题,那么就可以借助 于Ma亡lab,Ma艺heoa亡坛Ca, MaPI。等数学软件的最优化工具箱中的特定函数来求得其最优解.首先,我们要考虑参数a的最小二乘估计d的强相合性.若设d任0。为未知的参数真值,即往证尸(d一动=L 引理1.2.设时间序列{Xt}服从模型自.从E对<十co,那么可以选取参数空间为。1={o三a。三M,0三al+…+a。兰1,a‘)0江=1,2,…,g},其中M为吉林大学硕士学位论文序约未下ARCH模型最小二乘枯计3 充分大的正常数. 引理L3.设时间序列{瓜}服从模型(1 .l),E对<+oo,那么对撇。el,d是 EL。(a)关于。的唯一机小值点.其中d为参数真值. 引理1.4.假设时间序列{x‘}服从模型(l .l),E对<+oo,。。el,那么当 几*+co时,有 几=sup}L。(。)一EL。(a)】一0,a.s. a任el 定理1.1.设时间序列{瓜}服从模型(l .l),E对<+co,d任e0,那么 P{,溉d二定理1.2.如果定理忍.1的条件成立,d}二1d任eg,那么当n、+co时,有 .02L。(a、口ZL。(a、.suP}二一活:一二一石下,任六二}一U,a·s.U簇i,j簇q,。产00 aa咬U口布U口健aa布“仁Ul .J口JnZ rl丈八几2 1 1 v.八u习n、“)。ru叶气八忿一1,“)1一r了,、飞石石歹.一钊一万丽忑产-」}。一;=,\a)其中。2是el的内.点集. 定理1.3.假设时间序列{Xt}服从模型(l .l),E对<+co,d。创,那么双动>0. 引理1.5.假设时间序列{Xt}满足模型户.1),d。创,且E对<+co,则有去象一去豁[xl一h‘(xt一,;。)]2}a一。么万(o,。2其中沪=E姚军. 下面研究d的渐近正态性, 定理L4.假设时间序列{xt}满足模型自.l),d〔eg,那么有而(d一司乙N(o,I一‘(d扮21一‘(d)),其中刀2=E姚不.吉林大学硕士学位论文序约束下ARCH模型最小二乘估计4 下面讨论ARcH(0,q)模型参数序关系的检验问题.一般的,对于当前时刻t来说,延迟小的数据对现在的影响比延迟大的要大或至少相等,故要求系数满足一定约束的ARCH模型更切合实际,即参数满足序关系al)QZ)…)a中假设时间序列{凡}满足模型(1.1),d。创,考虑下面的检验问题H0:傲二a2=一=峋v.s.HI一H0·其中Hl:al)aZ)…)峋记C={a:Ql)aZ)…)a。},犷=(鱿,鱿,…,弓)T是参数a在约束集合c门e。上的最小二乘估计,集合C可以改写成 C={a:a了a)o,么=1,2,…,q一}, 其中。‘一(0…01廿呈。…0)T,、一1,2,…,。一1. 假设J是集合{1,2,…,q一l}的子集.记 BJ={a:叮a=0,坛。大叮a>0,乞贾去a任几。}, 力J={a:叮a二0,乞任J;lla一训(句}, AJ={a:好a二0,乞。J; a 0 eo}, 如果把判别函数L。(a)在AJ上关于。的最小值点记为d了,下面定理将给 出犷与d,应,之间的关系. 定理LS.对于模型(l .1),
【Abstract】 Autoregressive Conditional Heteroskedasticity (ARCH) model is very important in the study of nonlinear time series. In fact, the wave of stock price sequence is one focus of the money market and heteroskedasticity is impersonally existing phenomenon. In order to study the phemomenon actually, this model was first mentioned by Engle(1982) and developed by Engle and Kraft(1983). In this paper, we discuss a special case of ARCH model, namely ARCH(0, q).If a time series {Xt} satisfywhere and is independent with . The {Xt} is said to be ARCH(p, q).In this paper, we only discuss the stationary and ergodic ARCH(0, q) time series, namely {Xt} satisfyingwhere , , and t is independent with {Xs, s < t}. Let denote the o-field generated by {Xt-1,Xt-2...}.Prom thatwe study the functionwhereThen we will study the properties of the Least Square Estimate that minimizes Ln(a) in 0.Note a is the LSE of parameter a. a can be solved by = 0. And we can compute the LSE through the Quadratic Programming method.Matlab, Mathematica, Maple are all suitly solve the minimum.Firstly, we study the strong consistency of LSE d of the parameter a. Assume is the unknown true parameter, we should proov Lemma 1.1 Assuem that the time series {Xt} satisfies (1.1),EX4t < +, then we can use the parameter space , where M is a fixed large positive number.Lemma 1.2 Assume the time series {Xt} satisfies (1.1),EX4t < +, then for is the the unique minimum of ELn(a).Lemma 1.3 Assume the time series {Xt} satisfies (1.1),EX4t < + ,thenTheorem 1.1 Assume the time series {Xt} satisfies (1.1), EXf < +, a 0, thenTheorem 1.2 If the conditions of Theorem 1.1, thenwhen n , where the inner point set of .Theorem 1.3 Assume the time series {Xt} satisfies (1.1), EXf < +, a 10, then I (a) >0.Lemma 1.4 Assume the time senes {Xt} satisfies (1.1),a 02, and EX8t < +, thenFollowing , we study the asymptotic normality of a.Theorem 1.4 Assume the time series {Xt} satisfies (1.1),a 01, then n(a - Next, we consider the testing problem with order of ARCH(0,q). In many eco-nomical and financial studies, we expect that the effect of recent data is greater than that of remote data. This can be expressed asnamely parameters are restricted by a simple order.Suppose the time series {Xt} follows (1.1),a 0, consider the following testwhere Write the LSE of a in the restricted set C The set C can be rewrited aswhere Assume J is the subset of If write aJ as the minimum of Ln(a) in AJ, we can find the relationship of a* anda, aJ.Theorem 1.5 For (1.1), if and H\ hold, then from the large sample of view, there exists a J(J depends on w and n), we have a* = aJ. Namely, for almost Prom the large sample of view, we suppose : if , then Based on this supposition, we have the following theorem.Theorem 1.6 the conditions of Theorem 1.5 hold, then a {1,2, ... ,q -1} if and only if aJ satisfy the following conditions: Let satisfy i),ii),iii),}, when n is large enough,Thus, we obtain the exact expression of a*.Under H0, rewrite C0:Subsequently, we derive a representation of aJ by score function under H0. DefineBased on the above supposition, we obtainLemma 1.5 Under Ho, for each J {1,2, ... , q - 1},whereis the LSE of and ami is No. mi part of a. Andwhere denotes the limited distributions of nXn and Yn are same"From Lemma 1.5, for each JFurthermore,Let is the project matrix, and it is continuous. Sowhere Theorem 1.6 Under Ho,where k = q + 1 - rank(PJ).Lei AUJ is the set which composed with Uj satisfying 1,2,3?LetTheorem 1.7 Under H0,Theorem 1.8 Under H0, for t > 0,The results that we obtain for ARCH can be easily extended to B-ARCH with the similar score function , because ARCH is a special case of B-ARCH. And we omit the proofs.
- 【网络出版投稿人】 吉林大学 【网络出版年期】2004年 04期
- 【分类号】O211
- 【被引频次】1
- 【下载频次】213