Pure Mathematics
Vol. 08  No. 06 ( 2018 ), Article ID: 27470 , 5 pages
10.12677/PM.2018.86084

Upper Bounds of Moderate Deviations for the Estimator in the Non-Stationary Ornstein-Ulenbeck Process

Jin Shao

Department of Mathematics, Nanjing University of Aeronautics and Astronautics, Nanjing Jiangsu

Received: Oct. 15th, 2018; accepted: Oct. 27th, 2018; published: Nov. 8th, 2018

ABSTRACT

We study the maximum likelihood estimator of the drift estimation in a non-stationary Ornstein-Uhlenbeck process. Upper bounds of moderate deviations for this estimator are obtained.

Keywords:Drift Estimation, Moderate Deviations, Non-Stationary Ornstein-Uhleneck Process

非平稳Ornstein-Uhlenbeck过程中参数估计量的中偏差上界

邵 金

南京航空航天大学理学院数学系,江苏 南京

收稿日期:2018年10月15日;录用日期:2018年10月27日;发布日期:2018年11月8日

摘 要

对于非平稳Ornstein-Uhlenbeck过程,我们研究它的漂移项参数的极大似然估计量,得到了该估计量的中偏差上界。

关键词 :漂移项参数,中偏差,非平稳Ornstein-Uhlenbeck过程

Copyright © 2018 by author and Hans Publishers Inc.

This work is licensed under the Creative Commons Attribution International License (CC BY).

http://creativecommons.org/licenses/by/4.0/

1. 引言

考虑如下的非平稳Ornstein-Uhlenbeck (O-U)过程:

d X t = θ X t d t + d W t , X 0 = x 0 , t 0 (1)

其中W为标准布朗运动,参数 θ 未知。 P θ 表示(1)的解的概率分布,(1)的似然率过程可具体表示如下 [1] :

d P θ 1 d P θ 0 | F T = exp { ( θ 1 θ 0 ) 0 T X s d X s θ 1 2 θ 0 2 2 0 T X s 2 d s } (2)

其中, F T = σ ( W s , s T ) θ 0 , θ 1 。基于 { X t , t 0 } 的观测值, θ P θ 之下的极大似然估计量(MLE)为:

θ ^ T = 0 T X s d X s 0 T X s 2 d s .

已知 θ ^ T 是强相合的,但根据 θ 的值可知,分布行为和相应的速度是不同的。

1) 若 θ < 0 ,(1)中过程X是遍历的,且

T ( θ ^ T θ ) N ( 0 , 2 θ ) ,

其中 表示依分布收敛。Florens-Landais和Pham [2] 利用Gärtner-Ellis定理得到了大偏差。Bercu和Rouault [3] 提出了精细大偏差,而Guillin和Liptser [4] 得到了中偏差。Gao和Jiang [5] 研究了一些偏差不等式以及中偏差。

2) 若 θ > 0 ,(1)中过程X是非常返的,且

e θ T 2 θ ( θ ^ T θ ) ν η ,

其中, ν η 为两个独立的高斯随机变量 [6] 。

对于非平稳Ornstein-Uhlenbeck过程,如 θ > 0 的情况,Bercu,Coutin和Savy [7] 已经研究了 θ ^ T 的精细大偏差。本文受非平稳高斯自回归过程的中偏差启发,考虑估计量 θ ^ T 的中偏差上界。

2. 引理及证明

接下来介绍两个关键引理。

b T , T 0 为一个非负函数且满足 b T = ο ( T )

引理1:若 θ > 0 ,对任意的 α > 0 x ,我们有

lim T 1 b T log E P θ exp { α e 2 b T x 2 T θ 0 T X t 2 d t } = { x , x 0 ; 0 , x < 0.

证明:由Girsanov’s公式,对 μ > 0 ,Florens-Landais和Pham [2] 得到

log E P θ exp { μ 0 T X s 2 d s } = T 2 ( θ + θ 2 + 2 μ ) 1 2 log ( 1 2 θ 2 θ 2 + 2 μ + θ + θ 2 + 2 μ 2 θ 2 + 2 μ e 2 T θ 2 + 2 μ ) x 0 2 μ θ θ 2 + 2 μ coth ( T θ 2 + 2 μ )

θ > 0 时,有

log E P θ exp { α e 2 b T x 2 T θ 0 T X t 2 d t } = 1 2 log ( 1 2 θ 2 θ 2 + 2 α e 2 b T x 2 T θ + θ + θ 2 + 2 α e 2 b T x 2 T θ 2 θ 2 + 2 α e 2 b T x 2 T θ e 2 T θ 2 + 2 α e 2 b T x 2 T θ ) x 0 2 α e 2 b T x 2 T θ θ θ 2 + 2 α e 2 b T x 2 T θ coth ( T θ 2 + 2 α e 2 b T x 2 T θ ) T 2 ( θ + θ 2 + 2 α e 2 b T x 2 T θ ) : = L 1 ( T ) + L 2 ( T ) + L 3 ( T )

由泰勒公式,我们有

θ 2 + 2 α e 2 b T x 2 T θ = θ + α θ e 2 b T x 2 T θ + ο ( e 2 b T x 2 T θ ) ,

由此可推得

lim T L 2 ( T ) = θ x 0 2 , L 3 ( T ) = θ T + Ο ( T e 2 b T x 2 θ T ) . (3)

L 1 ( T ) ,由简单计算得到

θ 2 θ 2 + 2 α e 2 b T x 2 T θ = 1 2 1 + 2 α e 2 b T x 2 T θ θ 2 = 1 2 α 2 θ 2 e 2 b T x 2 T θ + ο ( e 2 b T x 2 T θ )

θ + θ 2 + 2 α e 2 b T x 2 T θ 2 θ 2 + 2 α e 2 b T x 2 T θ e 2 T θ 2 + 2 α e 2 b T x 2 T θ = e 2 T θ 2 + 2 α e 2 b T x 2 T θ + ο ( e 2 b T x 2 T θ ) .

因此,

L 1 ( T ) = 1 2 log ( α 2 θ 2 e 2 b T x 2 T θ + e 2 T θ 2 + 2 α e 2 b T x 2 T θ + ο ( e 2 b T x 2 T θ ) ) . (4)

x 0 ,则

lim T e 2 b T x 2 T θ e 2 T θ 2 + 2 α e 2 b T x 2 T θ = +

利用(3)和(4),可推得

lim T 1 b T ( L 1 ( T ) + L 3 ( T ) ) = lim T 1 b T ( 1 2 log ( α 2 θ 2 e 2 b T x 2 T θ + ο ( e 2 b T x 2 T θ ) ) θ T + Ο ( T e 2 b T x 2 T θ ) ) = x (5)

另一方面,若 x < 0 ,则

lim T e 2 b T x 2 T θ e 2 T θ 2 + 2 α e 2 b T x 2 T θ = 0 ,

可推得

lim T 1 b T ( L 1 ( T ) + L 3 ( T ) ) = lim T 1 b T ( 1 2 log ( e 2 T θ 2 + 2 α e 2 b T x 2 T θ + ο ( e 2 b T x 2 T θ ) ) θ T + Ο ( T e 2 b T x 2 T θ ) ) = 0 (6)

结合(3),(5)和(6),引理1得证。

引理2:对任意 r > 0 ,有

P θ ( | θ ^ T θ | r ) 2 inf q > 1 ( E P θ exp { r 2 2 ( q 1 ) 0 T X s 2 d s } ) 1 q .

证明:因为对任意 p > 1 ,有

exp { λ p 0 T X s d W s p 2 2 λ 2 0 T X s 2 d s } , T 0

F T -鞅,对任意 λ > 0

P ( θ ^ T θ r ) E P θ exp { λ 0 T X s d W s r λ 0 T X s 2 d s } = E P θ exp { λ 0 T X s d W s p 2 λ 2 0 T X s 2 d s + p 2 λ 2 0 T X s 2 d s r λ 0 T X s 2 d s } ( E P θ exp { λ p 0 T X s d W s p 2 2 λ 2 0 T X s 2 d s } ) 1 p ( E P θ exp { ( p q 2 λ 2 q r λ ) 0 T X s 2 d s } ) 1 q = ( E P θ exp { ( p q 2 λ 2 q r λ ) 0 T X s 2 d s } ) 1 q

其中, 1 p + 1 q = 1 。结合 inf λ > 0 { p q 2 λ 2 q r λ } = r 2 2 ( q 1 ) ,可得

P θ ( θ ^ T θ r ) inf q > 1 ( E P θ exp { r 2 2 ( q 1 ) 0 T X s 2 d s } ) 1 q .

故引理2得证。

3. 主要结论及证明

定理:若 θ > 0 { | e θ T ( θ ^ T θ ) | 1 b T , T > 0 } 以速度 b T 满足中偏差上界,且速率函数为

I ( x ) = { log x , x 1 ; 0 , 0 < x < 1 ; + , x 0.

如,对任意闭集 F

lim sup T 1 b T log P θ ( | e θ T ( θ ^ T θ ) | 1 b T F ) inf x F I ( x ) .

证明:对任意给定 x > 0 ,由引理2有

lim sup T 1 b T log P θ ( | e θ T ( θ ^ T θ ) | 1 b T x ) = lim sup T 1 b T log P θ ( log | e θ T ( θ ^ T θ ) | b T log x ) = lim sup T 1 b T log P θ ( | θ ^ T θ | e b T log x θ T ) lim sup T 1 b T inf q > 1 1 q log ( 2 E P θ exp { q 1 2 e 2 b T log x 2 θ T 0 T X s 2 d s } ) = sup q > 1 log x q = { log x , x 1 ; 0 , 0 < x < 1.

结合Worms [8] 中引理3,定理得证。

基金项目

南京航空航天大学2017年研究生创新基地(实验室)开放基金立项资助项目(项目编号:kfjj20170805)。

文章引用

邵 金. 非平稳Ornstein-Uhlenbeck过程中参数估计量的中偏差上界
Upper Bounds of Moderate Deviations for the Estimator in the Non-Stationary Ornstein-Ulenbeck Process[J]. 理论数学, 2018, 08(06): 632-636. https://doi.org/10.12677/PM.2018.86084

参考文献

  1. 1. Kutoyants, Yu.A. (2003) Statistical Inference for Ergodic Diffusion Process. Springer, Berlin.

  2. 2. Florens-Landais, D. and Pham, H. (1999) Large Deviations in Estimate of an Ornstein-Uhlenbeck Model. Journal of Applied Probability, 36, 60-77.
    https://doi.org/10.1239/jap/1032374229

  3. 3. Bercu, B. and Rouault, A. (2002) Sharp Large Deviations for the Ornstein-Uhlenbeck Process. Theory of Probability and Its Application, 46, 1-19.
    https://doi.org/10.1137/S0040585X97978737

  4. 4. Guillin, A. and Liptser, R. (2006) Examples of Moderate Deviation Principles for Diffusion Processes. Discrete and Continuous Dynamical Systems, Series B, 6, 77-102.

  5. 5. Gao, F.Q. and Jiang, H. (2009) Deviation Inequalities and Moderate Deviations for Estimators of Parameters in an Ornstein-Uhlenbeck Process with Linear Drift. Electronic Communications in Probability, 14, 210-223.
    https://doi.org/10.1214/ECP.v14-1466

  6. 6. Dietz, H.M. and Kutoyants, Yu.A. (2003) Parameter Estimation for Some Non-Recurrent Solutions of SDE. Statist. Decisions, 21, 29-45.
    https://doi.org/10.1524/stnd.21.1.29.20321

  7. 7. Bercu, B., Coutin, L. and Savy, N. (2012) Sharp Large Deviations for the Non-Stationary Ornstein-Uhlenbeck Process. Stochastic Processes and Their Applications, 122, 3393-3424.
    https://doi.org/10.1016/j.spa.2012.06.006

  8. 8. Worms, J. (2001) Large and Moderate Deviations Upper Bounds for the Gaussian Autoregressive Process. Statistics and Probability Letters, 51, 235-243.
    https://doi.org/10.1016/S0167-7152(00)00134-6

期刊菜单