On Finite Time Span Estimators of Parameters for Ornstein-Uhlenbeck Processes

Jun S. Han, Nino Kordzakhia School of Mathematical and Physical Sciences, Macquarie University (email: [email protected], [email protected]) · 2025

Read On Finite Time Span Estimators of Parameters for Ornstein-Uhlenbeck Processes by Jun S. Han, Nino Kordzakhia School of Mathematical and Physical Sciences, Macquarie University (email: [email protected], [email protected]) (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-23677.txt.

Introduction

In a 2025 arXiv paper, Jun S. Han and Nino Kordzakhia study guess errors for a mean-pulling process over short spans. The process, named for Ornstein and Uhlenbeck, drifts back toward a center value over time. Best-fit guesses of its traits are off in known ways when the watch span is short. The pair works out formulas for the bias and the squared error of those guesses.

Their math leans on a change of measure and on Ito's rule for random paths. A joint tool then gives numbers for the full process case. Charts show how bias and error move with the pull-back speed, the trait users care most about.

Catalog record

CREATORJun S. Han, Nino Kordzakhia School of Mathematical and Physical Sciences, Macquarie University (email: [email protected], [email protected])
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE52 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-23677

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

On Finite Time Span Estimators of Parameters for Ornstein-Uhlenbeck Processes Jun S. Han, Nino Kordzakhia School of Mathematical and Physical Sciences, Macquarie University (email: [email protected], [email protected]) Abstract We study the bias and the mean-squared error of the maximum likelihood estimators (MLE) of parameters associated with a two-parameter mean-reverting process for a finite time T T . Using the likelihood ratio process, we derive the expressions for MLEs, then compute the bias and the MSE via the change of measure and Ito’s formula. We apply the derived expressions to the general Ornstein-Uhlenbeck process, where the bias and the MSE are numerically computed through a joint moment-generating function of key functionals of the O–U process. A numerical study is provided to illustrate the behaviour of bias and the MSE for the MLE of the mean-reverting speed parameter. 1 Introduction We focus on the univariate two-parameter mean-reverting process \mathbf{Y}=\{Y_{t},0\leq t\leq T\} , which is defined through the following stochastic differential equation (SDE), dY_{t}=\lambda(\alpha-b_{t})dt+\sigma_{t}dW_{t},\quad Y_{0}=y,\quad\lambda\in{\Lambda},\quad\alpha\in\mathcal{A}. (1) The two parameters \lambda and \alpha are to be estimated from the following sets of observations \{Y_{t},b_{t},\sigma_{t},\;0\leq t\leq T\} in open intervals \Lambda\in\mathbb{R} and \mathcal{A}\in\mathbb{R} , respectively. The standard Brownian motion \mathbf{W}=\{W_{t},t\geq 0\} is defined on a probability space (\Omega,~\mathbf{F}^{W}=\{F_{t}^{W}\}_{t\geq 0},~\mathbb{P}) , with \mathbf{F}^{W} being a filtration generated by \mathbf{W} . b_{t} and \sigma_{t} are measurable processes with respect to \mathbf{F}^{Y}=\{F_{t}^{Y},~0\leq t\leq T\} , where F_{t}^{Y}={\mathcal{F}}\{Y_{s},0\leq s\leq t\} is the family of sigma-algebras generated by \mathbf{Y} . The stochastic process driven by SDE in ( 1 ) appears frequently in mathematical finance. For example, if b_{t}=Y_{t} and \sigma_{t}=\sigma , then Y becomes an Ornstein-Uhlenbeck process, which was introduced in finance in [ 41 ] . The work was extended to more general type of SDE in [ 7 ] , known as the CKLS (Chan-Karolyi-Longstaff-Sanders) process, with b_{t}=Y_{t} and \sigma_{t}=\sigma Y_{t}^{\gamma} , which is defined as dY_{t}=\lambda(\alpha-Y_{t})dt+\sigma Y_{t}^{\gamma}dW_{t},\quad 0\leq t\leq T. (2) Note that from ( 2 ), if \gamma=0 and \sigma_{t}=\sqrt{|Y_{t}|} , then this leads to a CIR (Cox-Ingersoll-Ross) process in [ 9 ] . These stochastic processes were widely used for modelling the term structure of interest rates, and for volatility estimation in a finite time span; see [ 35 ] and [ 21 ] . Several studies were made based on the SDE in ( 1 ). Theoretical results on the existence of strong and weak solutions were studied in the monographs in [ 23 ] and [ 24 ] , and statistical inferences for stochastic models described by SDE driven by Brownian motion have been presented in the monographs by [ 1 ] , [ 36 ] , [ 17 ] , [ 23 ] , [ 4 ] , [ 18 ] , [ 39 ] and [ 13 ] . The asymptotic properties of the estimators for discretised SDE-driven process in ( 1 ) under various discretisation schemes can be found in [ 14 ] , while the asymptotic properties of sequential estimators for discretised O-U process have been studied in [ 20 ] , [ 22 ] , and [ 32 ] . In continuous time, the asymptotic properties of the maximum likelihood estimators (MLEs) for the CKLS process in ( 2 ) were studied in [ 38 ] , [ 14 ] , [ 6 ] , [ 31 ] , [ 16 ] , [ 26 ] . In addition, the computation and asymptotic properties of the pseudo moment estimators were studied in [ 27 ] , while a least squares approach has been studied in [ 12 ] for a fractional Ornstein-Uhlenbeck processes, later extended to non-ergodic O–U processes in [ 3 ] , [ 10 ] and [ 11 ] . In this paper, we study the bias and the mean-squared error (MSE) of the maximum likelihood estimators \theta_{T} of \theta=(\lambda,\alpha) , which are defined as bias(\theta_{T})=\mathbb{E}_{\lambda,\alpha}[(\theta_{T}-\theta)],\quad mse(\theta_{T})=\mathbb{E}_{\lambda,\alpha}[(\theta_{T}-\theta)^{2}], with the assumption that the sample size T is fixed. Studies on bias of estimators in autoregressive processes began from the classical study of the estimation of the correlation coefficient. The analytical calculation of the bias in the correlation estimator was first explored in [ 25 ] and [ 15 ] in a stationary AR(1) model. In particular, [ 15 ] highlighted the limitations of asymptotic expansions when \rho is near unity, due to the skewness in the distribution of the estimator and the slow decay of higher-order terms. These concerns suggest that alternative approaches may be necessary to improve the accuracy of bias corrections. For a Gaussian stationary AR(1) process, the overview of asymptotics of least squares estimators for three parameters, including the bias corrections for the slope parameter was provided in [ 33 ] . These studies have triggered further developments for bias analysis of various types of estimators, including the maximum likelihood estimators (see [ 37 ] , [ 8 ] ). In the case of continuous time, the asymptotic expansions for the bias and MSE of the maximum likelihood estimators analytically for the SDE in ( 1 ), in particular for O–U and CIR processes, have been studied in [ 40 ] . The computational challenges arising from the dependence of the mean reversion parameter estimator on the initial value and discretisation frequency, as well as its slow convergence for small parameter values, have been examined in [ 2 ] . We will derive an integral transform using the change of measure and Ito’s formula to compute bias(\theta_{T}) and mse(\theta_{T}) for fixed T under the assumptions made in ( 1 ). This approach allows us to derive expansions for cases as \lambda\rightarrow 0 or at \lambda=\frac{u}{T} , where u is a constant. When referring to the study in [ 32 ] , the first-order asymptotic approximations for bias(\theta_{T}) and mse(\theta_{T}) could be misleading for moderate values of T or small values of \lambda . The bias of the MLE can be used in the bias correction to reduce the mean-squared error of the MLE. This paper is organised as follows. In Section 2 , we formulate some general results studied in [ 32 ] that are valid within the framework of the mean-reverting process in ( 1 ). These results provides the basis for studying the fixed sample properties of the MLE for the Ornstein-Uhlenbeck (O-U) process. In Section 3 , we study the properties of MLEs to provide the analytical formulae and numerical illustrations for the bias and the MSE of the MLEs, and its modification using the moment generating functions (MGFs). The derivations of these MGFs are included in the Appendix. In Section 4 , we provide a numerical study to illustrate the behaviour of bias and MSE for a finite sample size T . Section 5 concludes this paper. 2 Maximum Likelihood Estimators In this section, we derive the expressions of the MLEs for the two parameters \lambda and \alpha , and discuss their properties. 2.1 Notations and Assumptions We assume \mathbf{P}=\left\{\mathbb{P}_{\lambda,\alpha}(\alpha),A\in F_{T}^{Y}\right\} be the probability measure generated by the process Y in ( 1 ), with \mathbb{E}_{\lambda,\alpha} representing the expectation with respect to \mathbb{P}_{\lambda,\alpha} . In addition, assume that for any \lambda\in\Lambda and \alpha\in\mathcal{A} , where \Lambda and \mathcal{A} are open intervals in \mathbb{R} , the following conditions hold: \displaystyle\mathbb{P}_{\lambda,\alpha}\left\{\int_{0}^{T}|b_{t}|\;dt<\infty\right\}=\mathbb{P}_{\lambda,\alpha}\left\{\int_{0}^{T}\sigma_{t}^{2}\;dt<\infty\right\}=1, (3)

Read the complete text