Hostname: page-component-78c5997874-t5tsf Total loading time: 0 Render date: 2024-11-08T13:24:28.227Z Has data issue: false hasContentIssue false

An exponential nonuniform Berry–Esseen bound of the maximum likelihood estimator in a Jacobi process

Published online by Cambridge University Press:  14 February 2024

Hui Jiang*
Affiliation:
Nanjing University of Aeronautics and Astronautics
Qihao Lin*
Affiliation:
Nanjing University of Aeronautics and Astronautics
Shaochen Wang*
Affiliation:
South China University of Technology
*
*Postal address: School of Mathematics, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China.
*Postal address: School of Mathematics, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China.
**Postal address: School of Mathematics, South China University of Technology, Guangzhou 510640, China. Email: [email protected]
Rights & Permissions [Opens in a new window]

Abstract

We establish the exponential nonuniform Berry–Esseen bound for the maximum likelihood estimator of unknown drift parameter in an ultraspherical Jacobi process using the change of measure method and precise asymptotic analysis techniques. As applications, the optimal uniform Berry–Esseen bound and optimal Cramér-type moderate deviation for the corresponding maximum likelihood estimator are obtained.

Type
Original Article
Copyright
© The Author(s), 2024. Published by Cambridge University Press on behalf of Applied Probability Trust

1. Introduction and main results

1.1. Introduction

Consider the following ultraspherical Jacobi process

(1.1) \begin{equation} \textrm{d}{X_t} = bX_t\,\textrm{d}{t} + \sqrt{1-X_t^2}\,\textrm{d}{W_t}, \qquad X_0=0,\ t\geq0,\end{equation}

where $W=\{W_t\colon t\geq0\}$ is a standard Brownian motion with unknown drift parameters $b \in (\!-\!\infty,-1)$ . This process is well defined under the assumption $b<-1$ , i.e. we have $|X_t|<1$ for all $t\geq 0$ . The Jacobi process, also called Wright–Fisher diffusion, was originally used to model gene frequencies [Reference Ethier and Kurtz9, Reference Karlin and Taylor16]. More recently, it has also been applied to describe financial factors. For example, [Reference Delbane and Shirakawa7] models interest rates by the Jacobi process and studies moment-based techniques for pricing bonds. Moreover, this process has also been applied to model stochastic correlation matrices [Reference Ahdida and Alfonsi2] and credit default swap indexes [Reference Bernis and Scotti5]. For the multivariate case, see [Reference Ackerer, Filipović and Pulido1, Reference Gouriéroux and Jasiak13].

For $b \in (\!-\!\infty,-1)$ , the Jacobi process (1.1) has stationary distribution $\textrm{Beta}(\!-\!b,-b)$ , i.e. the Beta distribution with shape parameter $-b$ and scale parameter $-b$ . Here, b also represents the mean-reverting parameter. For practical applications, it is crucial to construct the estimator for b and study the corresponding asymptotic properties. Let $\mathbb{P}_b$ be the probability distribution of the solution of (1.1) on $C(\mathbb{R}_+,\mathbb{R})$ . Under $\mathbb{P}_b$ , the maximum likelihood estimator of b is given by

(1.2) \begin{equation} \widehat b_T = \frac{\int_{0}^{T}{X_t}/({1-X_t^2})\,\textrm{d}{X_t}}{\int_{0}^{T}{X_t^2}/({1-X_t^2}) \,\textrm{d}{t}}.\end{equation}

The Jacobi process was subordinated in [Reference Demni and Zani8] by the method of random time change, and the corresponding semi-group density was obtained. Together with the Girsanov formula technique [Reference Bercu and Richou4, Reference Florens-Landais and Pham12], they gave the large deviations for $\widehat b_T$ . For the Girsanov formula technique, see also [Reference Zang and Zhang20]. Moreover, [Reference Jiang, Gao and Zhao14] studied the moderate deviations for $\widehat b_T$ .

From [Reference Kutoyants17], it follows that

\begin{align*}\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b)\stackrel{\textrm{d}}{\longrightarrow}N(0,1),\end{align*}

which immediately implies that

(1.3) \begin{equation} \sup_{x\in\mathbb{R}}\bigg|\mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b)\leq x\bigg)-\Phi(x)\bigg|\to0\end{equation}

as $T\to\infty$ . The Berry–Esseen bound for (1.3) bounds the global convergence speed which is uniform in all $x\in\mathbb{R}$ . In this paper, our motivation is to study the nonuniform convergence rate of the difference

\begin{align*}\mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b)\leq x\bigg)-\Phi(x),\end{align*}

which will depend on x as well as T. This local bound provides more accurate information than the Berry–Esseen bound; see Corollaries 1.1 and 1.2 and Remark 1.1. Our approach is based on the change of measure method [Reference Wang and Jing19] and precise asymptotic analysis techniques [Reference Jiang and Zhou15]. The explicit calculations and estimations for the Laplace functionals of $\int_{0}^{T} {X_t}/({1-X_t^2}) \,\textrm{d}{X_t}$ and $\int_{0}^{T} {X_t^2}/({1-X_t^2}) \,\textrm{d}{t}$ play crucial roles. Moreover, the techniques used here could potentially be used for a wider scope of diffusions that have a structure similar to our model.

1.2. Main results

We now state the main results of this paper, i.e. the exponential nonuniform Berry–Esseen bound and its application for $\widehat b_T$ .

Theorem 1.1. There exists some positive constant $C>0$ depending only on b such that, for T large enough and any $\rho>0$ with $|x|<\rho T^{{1}/{6}}$ ,

(1.4) \begin{align} \bigg|\mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b)\leq x\bigg)-\Phi(x)\bigg| & \leq \frac{C}{\sqrt{T}\,}(x^2+1)\textrm{e}^{-{x^2}/{2}}, \end{align}
(1.5) \begin{align} \dfrac{\mathbb{P}_b(\sqrt{-{T}/({2(1+b)})}(\widehat b_T-b)>x)}{1-\Phi(x)} & = \exp\!\bigg\{O(1)\frac{x^3}{\sqrt{T}\,}\bigg\}, \end{align}

where the O(1) term only depends on b.

As applications of the above nonuniform Berry–Esseen bound, we can obtain the following optimal uniform Berry–Esseen bound and optimal Cramér-type moderate deviations for $\widehat b_T$ .

Corollary 1.1. There exists some positive constant $C>0$ depending only on b such that, for T large enough,

\begin{equation*} \sup_{x\in \mathbb{R}}\bigg|\mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b) \leq x\bigg) -\Phi(x)\bigg| \leq \frac{C}{\sqrt{T}\,}. \end{equation*}

Corollary 1.2. Let $\{\lambda_T,T>0\}$ be a family of positive numbers satisfying $\lambda_T/T^{{1}/{6}}\rightarrow 0$ , $T\rightarrow \infty$ . Then, for any $\rho\geq0$ ,

\begin{equation*} \sup_{0\leq x\leq \rho \lambda_T} \bigg|\dfrac{\mathbb{P}_b(\sqrt{-{T}/({2(1+b)})}(\widehat b_T-b)> x)}{1-\Phi(x)}-1\bigg| \rightarrow 0,\qquad T\to\infty. \end{equation*}

Remark 1.1. Note that, for any $x>0$ ,

\begin{align*} \frac{x}{\sqrt{2\pi}(1+x^2)}\exp\!\bigg\{{-}\frac{x^2}{2}\bigg\} \leq 1-\Phi(x) \leq \frac{1}{\sqrt{2\pi}x}\exp\!\bigg\{{-}\frac{x^2}{2}\bigg\}. \end{align*}

Hence, when $x\geq(\log T)^{1/2}$ , $1-\Phi(x)=o\big(T^{-1/2}\big)$ . Moreover, using [Reference Jiang, Gao and Zhao14, Theorem 1.1], for $x\geq(\log T)^{1/2}$ ,

\begin{align*} \mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b)>x\bigg) = o\big(T^{-1/2}\big). \end{align*}

Then,

\begin{align*} &\sup_{x\in\mathbb{R}}\bigg|\mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b) \leq x\bigg) - \Phi(x)\bigg| \\[5pt] & \quad = \sup_{|x|\leq(\log T)^{1/2} }\bigg|\mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b) \leq x\bigg) - \Phi(x)\bigg|. \end{align*}

Therefore, the uniform Berry–Esseen bound cannot characterize the error

\begin{align*} \mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b) \leq x\bigg) - \Phi(x) \end{align*}

for large x depending on T. Our nonuniform Berry–Esseen bound (1.4) obtained in Theorem 1.1) can fill this gap.

Remark 1.2. When T takes values in positive integers, letting

\begin{align*}\xi_i=\int_{i-1}^{i}\frac{X_t}{\sqrt{1-X_t^2}\,}\,\textrm{d}{W_t},\end{align*}

we have

\begin{equation*} \widehat{b}_T-b = \frac{\sum^{T}_{i=1}\xi_i}{\sum^{T}_{i=1}E(\xi_i^2\mid\mathcal{F}_{i-1})}, \end{equation*}

where $\mathcal{F}_{i}=\sigma(W_t, t\leq i)$ . Then $\{(\xi_i,\mathcal{F}_i), i\in\mathbb{N}\}$ is a sequence of martingale differences, and $\{\widehat{b}_T-b, T>0\}$ can be viewed as self-normalized martingales. The exponential nonuniform Berry–Esseen bound in Theorem 1.1 is a parallel result to the self-normalized sum of independent variables [Reference Wang and Jing19].

Moreover, it is worth noting that [Reference Fan and Shao11] obtained an exponential nonuniform Berry–Esseen bound and Cramér-type moderate deviation for self-normalized martingales. However, for the martingale differences $\{(\xi_i,\mathcal{F}_i), i\in\mathbb{N}\}$ it is difficult to verify the Bernstein condition in [Reference Fan and Shao11, (A1)]. So, our results cannot be covered or deduced directly by [Reference Fan and Shao11]. For more details on this topic, see [Reference Fan, Ion, Liu and Shao10] and the references therein.

The rest of this paper is organized as follows. In Section 2, we first give the explicit calculation of Laplace functionals related to a Jacobi process, which plays an important role in our asymptotic analysis. The proofs of the main results and their corollaries then follow in Section 3 and Appendix A.

2. Explicit calculation of Laplace functionals related to a Jacobi process

Recall from (1.2) that we can write

\begin{align*}\widehat b_T-b = \frac{\int_{0}^{T}{X_t}/({1-X_t^2})\,\textrm{d}{X_t} - b\int_{0}^{T}{X_t^2}/({1-X_t^2})\,\textrm{d}{t}} {\int_{0}^{T}{X_t^2}/({1-X_t^2})\,\textrm{d}{t}}.\end{align*}

Observe that, for $x>0$ ,

\begin{align*} & \mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b)> x\bigg) \\[5pt] &\quad = \mathbb{P}_b\bigg(\int_{0}^{T}\frac{X_t}{1-X_t^2}\,\textrm{d}{X_t} - \bigg(b+x\sqrt{-\frac{2(1+b)}{T}}\bigg)\int_{0}^{T}\frac{X_t^2}{1-X_t^2}\,\textrm{d}{t}>0\bigg),\end{align*}

and similarly for $x<0$ :

\begin{align*} & \mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b)\leq x\bigg)\\[5pt] &\quad = \mathbb{P}_b\bigg(\int_{0}^{T}\frac{X_t}{1-X_t^2}\,\textrm{d}{X_t} - \bigg(b+x\sqrt{-\frac{2(1+b)}{T}}\bigg)\int_{0}^{T}\frac{X_t^2}{1-X_t^2}\,\textrm{d}{t}\leq0\bigg).\end{align*}

Set

(2.1) \begin{equation} G_{T,x}=\int_{0}^{T}\frac{X_t}{1-X_t^2}\,\textrm{d}{X_t} - \bigg(b+x\sqrt{-\frac{2(1+b)}{T}}\bigg)\int_{0}^{T}\frac{X_t^2}{1-X_t^2}\,\textrm{d}{t} + x\sqrt{-\frac{T}{2(1+b)}};\end{equation}

then,

\begin{align*} \mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b)> x\bigg) & = \mathbb{P}_b\bigg(G_{T,x}>x\sqrt{-\frac{T}{2(1+b)}}\bigg), \qquad x>0, \\[5pt] \mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b)\leq x\bigg) & = \mathbb{P}_b\bigg(G_{T,x}\leq x\sqrt{-\frac{T}{2(1+b)}}\bigg), \qquad x < 0.\end{align*}

Write

(2.2) \begin{equation} h_{T,x}=x\sqrt{-\frac{2(b+1)}{T}},\end{equation}

and define a new probability measure $\mathbb{Q}$ which is absolutely continuous with respect to $\mathbb{P}_b$ , with Radon–Nikodym density

(2.3) \begin{equation} \frac{\textrm{d}\mathbb{Q}}{\textrm{d}\mathbb{P}_b} = \frac{\exp\{h_{T,x}G_{T,x}\}}{\mathbb{E}_b(\!\exp\{h_{T,x}G_{T,x}\})}.\end{equation}

Now, we can obtain, for $x>0$ ,

(2.4) \begin{align} & \mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b)> x\bigg) \nonumber \\[5pt] & \qquad\qquad = \mathbb{E}_\mathbb{Q}\bigg(\frac{\textrm{d}\mathbb{P}_b}{\textrm{d}\mathbb{Q}} \mathbf{1}_{\{G_{T,x}>x\sqrt{-{T}/({2(1+b)})}\}}\bigg) \nonumber \\[5pt] & \qquad\qquad = \mathbb{E}_b(\!\exp\{h_{T,x}G_{T,x}\})\mathbb{E}_\mathbb{Q} \Big(\!\exp\{-h_{T,x}G_{T,x}\}\mathbf{1}_{\{G_{T,x}>x\sqrt{-{T}/({2(1+b)})}\}}\Big), \end{align}

and, for $x<0$ ,

(2.5) \begin{align} & \mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b)\leq x\bigg) \nonumber \\[5pt] & \qquad\qquad = \mathbb{E}_b(\!\exp\{h_{T,x}G_{T,x}\})\mathbb{E}_\mathbb{Q} \Big(\!\exp\{-h_{T,x}G_{T,x}\}\mathbf{1}_{\{G_{T,x}\leq x\sqrt{-{T}/({2(1+b)})}\}}\Big). \end{align}

To prove Theorem 1.1, we need to analyze the Laplace functionals of $G_{T,x}$ under $\mathbb{P}_b$ and $\mathbb{Q}$ explicitly. Recall that if $\{X_t, t\geq 0\}$ is defined by (1.1), then its transition semi-group density is given by [Reference Demni and Zani8, p. 526]

(2.6) \begin{equation} P_t(y) = \sqrt{2\pi}K_{\alpha}\frac{\textrm{e}^{\gamma_0^2t/2}}{\sqrt{t}} \sum_{n=0}^{ \infty}\frac{\Gamma(2n+\alpha+\frac{3}{2})}{4^nn!\Gamma(n+\alpha+1)}(1-y^2)^{n+\alpha} f_{T_1}*f_{C_{2n+\gamma_0}}(1/t), \quad y\in [\!-\!1,1],\end{equation}

where $*$ is the convolution of two functions and

\begin{equation*} \alpha=-b-1, \quad K_{\alpha} = \frac{\Gamma(\alpha+1)}{\Gamma(\alpha+\frac{3}{2})2^{\alpha+{1}/{2}}}, \quad \gamma_0 = \alpha + \frac{1}{2}, \quad f_{T_1}(s) = \sum_{k\geq 0}\textrm{e}^{-({\pi^2}/{2})(k+{1}/{2})^2s}\mathbf{1}_{\{s>0\}},\end{equation*}

with

\begin{equation*} f_{C_{2n+\gamma_0}}(s) = \frac{2^{2n+\gamma_0}}{\sqrt{2\pi s^3}\Gamma (2n+\gamma_0)} \sum_{k\geq 0}(\!-\!1)^k\frac{(2k+2n+\gamma_0)\Gamma(k+2n+\gamma_0)}{k!} \textrm{e}^{-({(2k+2n+\gamma_0)^2})/{2s}}\mathbf{1}_{\{s>0\}}.\end{equation*}

The following lemma plays an important role in our analysis.

Lemma 2.1. For any $u\in\mathbb{R}$ and $h_{T,x}$ as defined in (2.2),

(2.7) \begin{multline} \mathbb{E}_b\exp\!\bigg\{\bigg(h_{T,x}+\frac{u\textrm{i}}{\sqrt{T}\,}\bigg)G_{T,x}\bigg\} \\[5pt] = \exp\!\bigg\{\bigg(h_{T,x}+\frac{u\textrm{i}}{\sqrt{T}}\bigg) \bigg(x\sqrt{-\frac{T}{2(1+b)}}-\frac{T}{2}\bigg) - (b-c_{T,x}(u))\frac{T}{2} + \mathcal{R}_{T,x}(u)\bigg\} \end{multline}

holds for all $T>0$ . Here, $\textrm{i}=\sqrt{-1}$ and

\begin{align*} c_{T,x}(u) & = -1 - \sqrt{(b+1)^2 + 2\bigg(h_{T,x}+\frac{\textrm{i}u}{\sqrt{T}}\bigg) \bigg(b+1+x\sqrt{-\frac{2(1+b)}{T}}\bigg)}, \\[5pt] \varphi_{T,x}(u) & = h_{T,x} + \frac{\textrm{i}u}{\sqrt{T}} + b - c_{T,x}(u), \\[5pt] \mathcal{R}_{T,x}(u) & = \log\Bigg(\sqrt{2\pi}K_{\alpha_1}\frac{\textrm{e}^{\gamma_1^2T/2}}{\sqrt{T}} \sum_{n=0}^{\infty}\frac{\Gamma(2n-c_{T,x}(u)+\frac{1}{2})}{4^nn!\Gamma(n-c_{T,x}(u))} f_{T_1}*f_{C_{2n+\gamma_1}}(1/T) \\[5pt] & \qquad\qquad \times B\bigg(n-c_{T,x}(u)-\frac{1}{2}\varphi_{T,x}(u),\frac{1}{2}\bigg)\Bigg). \end{align*}

Proof. From the definition of $G_{T,x}$ (2.1), it follows that

\begin{align*} & \mathbb{E}_b\exp\!\bigg\{\bigg(h_{T,x}+\frac{u\textrm{i}}{\sqrt{T}\,}\bigg)G_{T,x}\bigg\} \\[5pt] & \qquad = \exp\!\bigg\{\bigg(h_{T,x}+\frac{u\textrm{i}}{\sqrt{T}}\bigg) \bigg(x\sqrt{-\frac{T}{2(1+b)}}\bigg)\bigg\} \\[5pt] & \qquad\quad \times \mathbb{E}_b\exp\!\bigg\{\lambda_{T,x}(u)\int_{0}^{T}\frac{X_t}{1-X_t^2}\,\textrm{d}{X_t} - \mu_{T,x}(u)\int_{0}^{T}\frac{X_t^2}{1-X_t^2}\,\textrm{d}{t}\bigg\}, \end{align*}

where $\lambda_{T,x}(u)$ and $\mu_{T,x}(u)$ are defined by

\begin{equation*} \lambda_{T,x}(u) = h_{T,x} + \frac{u\textrm{i}}{\sqrt{T}}, \qquad \mu_{T,x}(u) = \bigg(h_{T,x} + \frac{u\textrm{i}}{\sqrt{T}}\bigg)\bigg(b + x\sqrt{-\frac{2(1+b)}{T}}\bigg). \end{equation*}

Let us define the joint log-Laplace transformation

\begin{equation*} g_T(\lambda_{T,x}(u),\mu_{T,x}(u)) = \log\mathbb{E}_b\exp\!\bigg\{\lambda_{T,x}(u)\int_{0}^{T}\frac{X_t}{1-X_t^2}\,\textrm{d}{X_t} - \mu_{T,x}(u)\int_{0}^{T}\frac{X_t^2}{1-X_t^2}\,\textrm{d}{t}\bigg\}. \end{equation*}

Denote the Brownian filtration by $\mathcal{F}_t=\sigma(W_s,s \leq t)$ ; then, by virtue of the Girsanov theorem, we have, for any $\textrm{Re}(b_0)<-1$ ,

\begin{equation*} \frac{\textrm{d}{\mathbb{P}_b}}{\textrm{d}{\mathbb{P}_{b_0}}}\bigg|_{\mathcal{F}_t} = \exp\!\bigg\{(b-b_0)\int_{0}^{t}\frac{X_s}{1-X_s^2}\,\textrm{d}{X_s} - \frac{1}{2}(b^2-b_0^2) \int_{0}^{t}\frac{X_s^2}{1-X_s^2}\,\textrm{d}{s}\bigg\}, \qquad t\geq0. \end{equation*}

Now, choose $b_0= -1 -\sqrt{(b+1)^2 +2(\lambda_{T,x}(u) +\mu_{T,x}(u))}\;:\!=\;c_{T,x}(u)\;:\!=\;c$ ; then

\begin{align*} &\exp\{ g_T(\lambda_{T,x}(u),\mu_{T,x}(u))\} \\[5pt] & \qquad = \mathbb{E}_c\bigg(\frac{\textrm{d}{\mathbb{P}_{b}}}{\textrm{d}{\mathbb{P}_{c}}} \exp\!\bigg\{\lambda_{T,x}(u)\int_{0}^{T}\frac{X_t}{1-X_t^2}\,\textrm{d}{X_t} - \mu_{T,x}(u)\int_{0}^{T}\frac{X_t^2}{1-X_t^2}\,\textrm{d}{t}\bigg\}\bigg) \\[5pt] & \qquad = \mathbb{E}_c\exp\!\bigg\{(\lambda_{T,x}(u)+b-c_{T,x}(u)) \int_{0}^{T}\frac{X_t}{1-X_t^2}\,\textrm{d}{X_t} \\[5pt] & \qquad\qquad\qquad\quad - \bigg(\mu_{T,x}(u) +\frac{1}{2}(b^2-c_{T,x}^2(u))\bigg) \int_{0}^{T}\frac{X_t^2}{1-X_t^2}\,\textrm{d}{t}\bigg\} \\[5pt] & \qquad = \mathbb{E}_c\exp\!\bigg\{(\lambda_{T,x}(u)+b-c_{T,x}(u)) \bigg(\int_{0}^{T}\frac{X_t}{1-X_t^2}\,\textrm{d}{X_t} + \int_{0}^{T}\frac{X_t^2}{1-X_t^2}\,\textrm{d}{t}\bigg)\bigg\}. \end{align*}

Applying Itô’s formula,

\begin{equation*} -\frac{1}{2}\log(1-X_T^2) - \frac{T}{2} = \int_{0}^{T}\frac{X_t}{1-X_t^2}\,\textrm{d}{X_t} + \int_{0}^{T}\frac{X_t^2}{1-X_t^2}\,\textrm{d}{t} \end{equation*}

and, as a consequence,

\begin{equation*} \exp\{g_T(\lambda_{T,x}(u),\mu_{T,x}(u))\} = \exp\!\bigg\{{-}\frac{T}{2}\varphi_{T,x}(u)\bigg\}\mathbb{E}_c(1-X_T^2)^{-\varphi_{T,x}(u)/2}. \end{equation*}

Using the transition density function of $X_{T}$ [Reference Demni and Zani8], see also (2.6), we obtain

\begin{align*} & \mathbb{E}_c(1 - X_T^2)^{-\varphi_{T,x}(u)/2} \\[5pt] & = \sqrt{2\pi}K_{\alpha_1}\frac{\textrm{e}^{\gamma_1^2{T}/{2}}}{\sqrt{T}} \sum_{n=0}^{\infty}\frac{\Gamma(2n-c_{T,x}(u)+\frac{1}{2})}{4^nn!\Gamma(n-c_{T,x}(u))} f_{T_1}*f_{C_{2n+\gamma_1}}(1/T)B\bigg(n-c_{T,x}(u)-\frac{1}{2}\varphi_{T,x}(u),\frac{1}{2}\bigg), \end{align*}

where B stands for the Beta function and

\begin{equation*} \alpha_1=-c_{T,x}(u)-1, \qquad K_{\alpha_1} = \frac{\Gamma(\alpha_1+1)}{\Gamma(\alpha_1+\frac{3}{2})2^{\alpha_1+{1}/{2}}}, \qquad \gamma_1=\alpha_1+\frac{1}{2}. \end{equation*}

Recalling the definition of $\mathcal{R}_{T,x}(u)$ in the above lemma we can write

\begin{align*} & f_{T_1}*f_{C_{2n+\gamma_1}}(1/T) \\[5pt] & = \sum_{k,\ell\geq0}\frac{(\!-\!1)^k}{\sqrt{2\pi}}U_{k,n}(u) \int_{T}^{\infty}\exp\!\bigg\{{-}\frac{1}{2}\bigg[(2n+2k+\gamma_1)^2s+\pi^2 \bigg(\ell+\frac{1}{2}\bigg)^2\bigg(\frac{s-T}{Ts}\bigg)\bigg]\bigg\}\frac{\textrm{d}{s}}{\sqrt{s}\,},\end{align*}

where

\begin{align*}U_{k,n}=\frac{\Gamma(2n+k+\gamma_1)2^{2n+\gamma_1}(2n+2k+\gamma_1)}{k!\Gamma(2n+\gamma_1)}.\end{align*}

In order to derive the precise estimation of $\mathcal{R}_{T,x}(u)$ , the following expansion of Gamma functions plays a crucial role.

Lemma 2.2. Let $\Gamma(z)$ be the Gamma function defined on the complex plane $\mathbb{C}$ . Then

\begin{equation*} \frac{\Gamma(z_2)}{\Gamma(z_1)} = \frac{z_1}{z_2}\exp\!\Bigg\{(z_1-z_2)\Bigg(\gamma-\sum_{k=1}^\infty\frac{z_2}{k(k+z_2)}\Bigg) + {O\bigg(\frac{|z_1-z_2|^2}{|z_2|^2}\bigg)}\Bigg\}, \end{equation*}

where $z_1,z_2\in \mathbb{C}$ , $\textrm{Re}(z_1), \textrm{Re}(z_2)>0$ , and $\gamma=0.5772\ldots$ is the Euler constant. Moreover, when $z_1-z_2=c\in\mathbb{R_+}$ and $\textrm{Re}(z_2)>1$ ,

\begin{equation*} \bigg|\frac{\Gamma(z_2)}{\Gamma(z_1)}\bigg| \leq \bigg|\frac{z_1}{z_2}\bigg| \exp\{c\gamma\}. \end{equation*}

Proof. By Euler’s formula [Reference Ahlfors3, p. 199] we have, for all $z\in\mathbb{C}$ ,

\begin{align*} \frac{1}{\Gamma(z)} = z \textrm{e}^{\gamma z}\prod_{k=1}^\infty\bigg(1+\frac{z}{k}\bigg)\textrm{e}^{-z/k}, \end{align*}

where $\gamma=0.5772\ldots$ is the Euler constant. As a consequence,

\begin{align*} \frac{\Gamma(z_2)}{\Gamma(z_1)} & = \frac{z_1}{z_2}\textrm{e}^{\gamma(z_1-z_2)}\prod_{k=1}^\infty\frac{k+z_1}{k+z_2}\textrm{e}^{-({z_1-z_2})/{k}} \\[5pt] & = \frac{z_1}{z_2}\exp\!\Bigg\{\gamma(z_1-z_2) + \sum_{k=1}^\infty-\frac{z_1-z_2}{k} + \sum_{k=1}^\infty\log\bigg(1+\frac{z_1 -z_2}{k+z_2}\bigg)\Bigg\} \\[5pt] & = \frac{z_1}{z_2}\exp\!\Bigg\{\gamma(z_1-z_2) + (z_1-z_2)\sum_{k=1}^\infty\frac{-z_2}{k(k+z_2)} + O\Bigg(\sum_{k=1}^\infty\bigg|\frac{z_1-z_2}{k+z_2}\bigg|^2\Bigg)\Bigg\} \\[5pt] & = \frac{z_1}{z_2}\exp\!\Bigg\{(z_1-z_2)\Bigg(\gamma - \sum_{k=1}^\infty\frac{z_2}{k(k+z_2)}\Bigg) + {O\bigg(\frac{|z_1-z_2|^2}{|z_2|^2}\bigg)}\Bigg\}. \end{align*}

Moreover, when $z_1-z_2=c\in\mathbb{R_+}$ and $\textrm{Re}(z_2)>0$ , we have

\begin{equation*} \Bigg|\prod_{k=1}^\infty\bigg(1 + \frac{c}{k+z_2}\bigg)\textrm{e}^{-{c}/{k}}\Bigg| \leq \prod_{k=1}^\infty\bigg(1+\frac{c}{k}\bigg)\textrm{e}^{-{c}/{k}} \leq \prod_{k=1}^\infty\textrm{e}^{{c}/{k}}\textrm{e}^{-{c}/{k}} \leq 1. \end{equation*}

Thus, we obtain the desired estimate,

\begin{equation*} \bigg|\frac{\Gamma(z_2)}{\Gamma(z_1)}\bigg| \leq \bigg|\frac{z_1}{z_2}\bigg|\exp\{c\gamma\}. \end{equation*}

Define the Laplace functional of $G_{T,x}$ under $\mathbb{Q}$ as

\begin{equation*} \varphi_{\mathbb{Q},G_{T,x}}(u) = \mathbb{E}_{\mathbb{Q}}\exp\{\textrm{i}uG_{T,x}\}, \qquad u\in \mathbb{R}.\end{equation*}

In the following propositions, we give explicit asymptotic expansions of $\varphi_{\mathbb{Q},G_{T,x}}({u}/{\sqrt{T}})$ for different ranges of u.

Proposition 2.1. For any constant $\rho>0$ , and $|u|\leq\rho T^{{1}/{6}}$ , $|x|\leq\rho T^{{1}/{6}}$ ,

(2.8) \begin{align} \mathbb{E}_b\exp\{h_{T,x}G_{T,x}\} & = \exp\!\bigg\{\frac{x^2}{2} + O(x^3T^{-{1}/{2}})\bigg\}, \end{align}
(2.9) \begin{align} \varphi_{\mathbb{Q},G_{T,x}}\bigg(\frac{u}{\sqrt{T}\,}\bigg) & = \exp\{\mathcal{A}_{1,T,x}u\textrm{i} + \mathcal{A}_{2,T,x}u^2 + O(u^3T^{-{1}/{2}})\}, \end{align}

where

\begin{align*} \mathcal{A}_{1,T,x} = x\sqrt{-\frac{1}{2(1+b)}} + O(x^2 T^{-{1}/{2}}), \qquad \mathcal{A}_{2,T,x} = \frac{1}{4(b+1)} + O(xT^{-{1}/{2}}). \end{align*}

Proof. By using (2.3) and (2.7), we can write

(2.10) \begin{equation} \mathbb{E}_b\exp\{h_{T,x}G_{T,x}\} =\exp\!\bigg\{h_{T,x}\bigg(x\sqrt{-\frac{T}{2(1+b)}}-\frac{T}{2}\bigg) - (b-c_{T,x}(0))\frac{T}{2} + \mathcal{R}_{T,x}(0)\bigg\} \end{equation}

and

(2.11) \begin{align} \varphi_{\mathbb{Q},G_{T,x}}\bigg(\frac{u}{\sqrt{T}\,}\bigg) & = \mathbb{E}_b\bigg(\frac{\textrm{d}\mathbb{Q}}{\textrm{d}\mathbb{P}_b} \exp{\bigg\{\frac{\textrm{i}u}{\sqrt{T}\,}G_{T,x}\bigg\}}\bigg) \nonumber \\[5pt] & = \frac{\mathbb{E}_b\exp{\{(h_{T,x} + {\textrm{i}u}/{\sqrt{T}})G_{T,x}\}}} {\mathbb{E}_b\exp{\{h_{T,x}G_{T,x}\}}} \nonumber \\[5pt] & = \exp\!\bigg\{\frac{u\textrm{i}}{\sqrt{T}\,}\bigg(x\sqrt{-\frac{T}{2(1+b)}}-\frac{T}{2}\bigg) - (c_{T,x}(0)-c_{T,x}(u))\frac{T}{2} + \mathcal{R}_{T,x}(u) - \mathcal{R}_{T,x}(0)\bigg\}. \end{align}

On the one hand, by Taylor’s expansion, straightforward but tedious calculations will give

(2.12) \begin{align} c_{T,x}(u) & = -1 + (b+1)\bigg(1 + 2\bigg(h_{T,x} + \frac{\textrm{i}u}{\sqrt{T}\,}\bigg) \bigg(\frac{1}{b+1} + \frac{x}{(b+1)^2}\sqrt{-\frac{2(1+b)}{T}}\bigg)\bigg)^{-{1}/{2}} \nonumber \\[5pt] & = \bigg(b + x\sqrt{-\frac{2(b+1)}{T}} - \frac{x^2}{T} + O(x^3T^{-{3}/{2}})\bigg) + (T^{-{1}/{2}} + O(x^2T^{-{3}/{2}}))u\textrm{i} \nonumber \\[5pt] & \quad + \bigg(\frac{1}{2T(b+1)} + O(xT^{-{3}/{2}})\bigg)u^2 + O(u^3T^{-{3}/{2}}), \end{align}
(2.13) \begin{align} c_{T,x}(0) & = b + x\sqrt{-\frac{2(b+1)}{T}} - \frac{x^2}{T} + O(x^3T^{-{3}/{2}}). \end{align}

Consequently, we find that

(2.14) \begin{align} -(c_{T,x}(0) - c_{T,x}(u))\frac{T}{2} & = \bigg(\frac{T^{{1}/{2}}}{2} + O(x^2T^{-{1}/{2}})\bigg)u\textrm{i}\nonumber \\[5pt] &\quad+ \bigg(\frac{1}{4(b+1)}+O(xT^{-{1}/{2}})\bigg)u^2 + O(u^3T^{-{1}/{2}}), \end{align}
(2.15) \begin{align} -(b - c_{T,x}(0))\frac{T}{2} & = \frac{x}{2}\sqrt{-2T(b+1)} - \frac{x^2}{2} + O(x^3T^{-{1}/{2}}). \end{align}

On the other hand, by Proposition A.1,

\begin{align*} \mathcal{R}_{T,x}(u) - \mathcal{R}_{T,x}(0) = O(T^{-{1}/{2}})u\textrm{i} + O(u^2T^{-1}) + O(u^3T^{-{3}/{2}}), \quad \mathcal{R}_{T,x}(0) = O(xT^{-{1}/{2}}). \end{align*}

Therefore, combining (2.10), (2.15), and (A.1), we get (2.8). Moreover, (2.9) can be obtained via (2.11), (2.14), and (A.1).

Proposition 2.2. For any constant $\rho>0$ , and $|u|\leq({|b|}/{32})T^{{1}/{2}}$ , $|x|\leq\rho T^{{1}/{6}}$ ,

\begin{equation*} \bigg|\varphi_{\mathbb{Q},G_{T,x}}\bigg(\frac{u}{\sqrt{T}\,}\bigg)\bigg| \leq 4(\sqrt{2} + 1)\sqrt{\pi}\exp\!\bigg\{{-}\frac{u^2}{8\sqrt{2}|b|} + \frac{\gamma}{2}\bigg\}, \end{equation*}

where $\gamma$ is the Euler constant.

Proof. Recall the calculation in (2.11); it suffices to analyze the real part of $-(c_{T,x}(0)-c_{T,x}(u))T/2$ and give an upper bound for $|\exp(\mathcal{R}_{T,x}(u))|$ . Using

(2.16) \begin{align} \textrm{Re}(\sqrt{1+z}) & = \sqrt{\frac{|1+z|+\textrm{Re}(1+z)}{2}}, c_{T,x}(u) \end{align}
(2.17) \begin{align} & = c_{T,x}(0)\bigg(1 + \frac{4u\textrm{i}}{c_{T,x}^2(0)\sqrt{T}\,} \bigg(b+1+x\sqrt{-\frac{2(1+b)}{T}}\bigg)\bigg)^{{1}/{2}}, \end{align}

we have

\begin{equation*} \textrm{Re}(c_{T,x}(u)) = \frac{c_{T,x}(0)}{\sqrt{2}}\Bigg(1 + \sqrt{1+\frac{16u^2}{c_{T,x}^4(0)T}\bigg(b+1+x\sqrt{-\frac{2(1+b)}{T}}\bigg)^2}\Bigg)^{{1}/{2}}. \end{equation*}

Now, for T large enough and $|u|\leq({|b|}/{32})T^{{1}/{2}}$ , $|x|\leq\rho T^{{1}/{6}}$ ,

\begin{equation*} -\frac{b}{2} \leq - b - 1 - x\sqrt{-\frac{2(1+b)}{T}} \leq -2b, \qquad \bigg|\frac{b}{2}\bigg| \leq |c_{T,x}(0)| \leq 2^{{1}/{4}}|b|. \end{equation*}

Consequently,

\begin{align*} & \frac{T}{2}|{\textrm{Re}(c_{T,x}(u)-c_{T,x}(0))}| \\[5pt] & = \frac{|c_{T,x}(0)|T}{2\sqrt{2}}\Bigg(\sqrt{1 + \bigg|1 + \frac{4u\textrm{i}}{c_{T,x}^2(0)\sqrt{T}\,} \bigg(b+1+x\sqrt{-\frac{2(1+b)}{T}}\bigg)\bigg|} - \sqrt{2}\Bigg) \\[5pt] & = \frac{4\sqrt{2}u^2(b+1+x\sqrt{-{2(1+b)}/{T}})^2}{|c_{T,x}^3(0)|} \bigg(\bigg|1+\frac{4u\textrm{i}}{c_{T,x}^2(0)\sqrt{T}\,}\bigg(b+1+x\sqrt{-\frac{2(1+b)}{T}}\bigg)\bigg|+1\bigg)^{-1}\\[5pt] & \quad \times \Bigg(\sqrt{1 + \bigg|1 + \frac{4u\textrm{i}}{c_{T,x}^2(0)\sqrt{T}\,} \bigg(b+1+x\sqrt{-\frac{2(1+b)}{T}}\bigg)\bigg|} + \sqrt{2}\Bigg)^{-1} \\[5pt] & \geq \frac{u^2(b+1+x\sqrt{-{2(1+b)}/{T}})^2}{|c_{T,x}^3(0)|} \bigg(1 + \frac{16u^2}{c_{T,x}^4(0)T}\bigg(b+1+x\sqrt{-\frac{2(1+b)}{T}}\bigg)^2\bigg)^{-{3}/{4}} \\[5pt] & \geq \frac{u^2}{8\sqrt{2}|b|}, \end{align*}

which immediately implies that

(2.18) \begin{equation} \textrm{Re}\bigg({-}(c_{T,x}(0)-c_{T,x}(u))\frac{T}{2}\bigg) \leq -\frac{u^2}{8\sqrt{2}|b|}. \end{equation}

Next, we turn to estimating $|\exp(\mathcal{R}_{T,x}(u))|$ . From (2.17),

\begin{equation*} \frac{|b|}{2} \leq |c_{T,x}(u)| = |c_{T,x}(0)| \bigg|1 + \frac{4u\textrm{i}}{c_{T,x}^2(0)\sqrt{T}\,}\bigg(b+1+x\sqrt{-\frac{2(1+b)}{T}}\bigg)\bigg|^{{1}/{2}} \leq \sqrt{2}|b|. \end{equation*}

Together with (2.12) and (A.2)–(A.4), we have

\begin{equation*} \bigg|{-}c_{T,x}(u) - \frac{1}{2}\varphi_{T,x}(u) + \frac{1}{2}\bigg| \leq (\sqrt{2} +1 )|b|, \qquad \bigg|{-}c_{T,x}(u) - \frac{1}{2}\varphi_{T,x}(u)\bigg| \geq \frac{1}{2}|b|. \end{equation*}

By Lemma 2.2, we have

\begin{equation*} \bigg|\frac{\Gamma(\!-\!c_{T,x}(u) - ({1}/{2})\varphi_{T,x}(u))} {\Gamma(\!-\!c_{T,x}(u) - ({1}/{2})\varphi_{T,x}(u) + {1}/{2})}\bigg| \leq \bigg|\frac{-c_{T,x}(u) - ({1}/{2})\varphi_{T,x}(u) + {1}/{2}}{-c_{T,x}(u) - ({1}/{2})\varphi_{T,x}(u)}\bigg| \textrm{e}^{{\gamma}/{2}} \leq 2(\sqrt{2} + 1)\textrm{e}^{{\gamma}/{2}}, \end{equation*}

which, together with (A.8), implies that

(2.19) \begin{equation} |\exp\{\mathcal{R}_{T,x}(u)\}| = \bigg|B\bigg({-}c_{T,x}(u) - \frac{1}{2}\varphi_{T,x}(u), \frac{1}{2}\bigg)(1 + O(\textrm{e}^{-2T}))\bigg| \leq 4(\sqrt{2} + 1)\sqrt{\pi}\textrm{e}^{{\gamma}/{2}}. \end{equation}

Finally, by (2.11), (2.18), and (2.19), we can complete the proof of this proposition.

3. Proofs of the main results

In this section we give the proof of our main result, Theorem 1.1, and its two corollaries. Recall that the definitions of $G_{T,x}$ and the probability measure $\mathbb{Q}$ are given in (2.1) and (2.3) respectively. Consider the normalized version of $G_{T,x}$ ,

\begin{equation*} \widetilde{G}_{T,x} = \bigg({-}\frac{T}{2(b+1)}\bigg)^{-{1}/{2}}\bigg(G_{T,x}-x\sqrt{-\frac{T}{2(b+1)}}\bigg).\end{equation*}

Using the simple identity $\int_{0}^{\infty}\exp\!{\{-xu\}}\,\textrm{d}{\Phi(u)} = \textrm{e}^{{x^2}/{2}}(1-\Phi(x))$ and (2.4), we can write, for $x>0$ ,

(3.1) \begin{align} \mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b)> x\bigg) & = \textrm{e}^{-x^2}\mathbb{E}_b(\!\exp\{h_{T,x}G_{T,x}\}) \int_{0}^{\infty}\exp\{-xt\}\,\textrm{d}{\mathbb{Q}(\widetilde{G}_{T,x}\leq t)} \nonumber \\[5pt] & = \textrm{e}^{-x^2}\mathbb{E}_b(\!\exp\{h_{T,x}G_{T,x}\})(\mathcal{J}_{T,x}+\textrm{e}^{{x^2}/{2}}(1-\Phi(x))),\end{align}

and for $x<0$ , using (2.5),

\begin{equation*} \mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b)\leq x\bigg) = \mathbb{E}_b(\!\exp\{h_{T,x}G_{T,x}\})(\widehat{\mathcal{J}}_{T,x} + \textrm{e}^{{x^2}/{2}}\Phi(x)),\end{equation*}

where $h_{T,x}=x\sqrt{-{2(b+1)}/{T}}$ and

(3.2) \begin{equation} \mathcal{J}_{T,x} = \int_{0}^{\infty}\textrm{e}^{-xt}\,\textrm{d}{(\mathbb{Q}(\widetilde{G}_{T,x}\leq t)-\Phi(t))}, \quad \widehat{\mathcal{J}}_{T,x} = \int_{-\infty}^{0}\textrm{e}^{-xt}\,\textrm{d}{(\mathbb{Q}(\widetilde{G}_{T,x}\leq t)-\Phi(t))}.\end{equation}

Next, we need to give the estimation of $\mathcal{J}_{T,x}$ and $\widehat{\mathcal{J}}_{T,x}$ . The following lemma [Reference Petrov18, Theorem 2, p. 109] will play a crucial role.

Lemma 3.1. Assume F and G are probability distribution functions, and that G has bounded derivative. Define $\varphi(u)=\int_{\mathbb{R}}\textrm{e}^{\textrm{i}ux}\,\textrm{d}{F(x)}$ and $\psi(u)=\int_{\mathbb{R}}\textrm{e}^{\textrm{i}ux}\,\textrm{d}{G(x)}$ . Then, for any $M>0$ ,

\begin{equation*} \sup_{x\in\mathbb{R}}|F(x)-G(x)| \leq \frac{1}{\pi}\int_{-M}^{M}\bigg|\frac{\varphi(u)-\psi(u)}{u}\bigg|\,\textrm{d}{u} + \frac{24}{\pi M}\sup_{x\in\mathbb{R}}|G'(x)|. \end{equation*}

Proof of Theorem 1.1. Without of loss of generality, we assume that $x>0$ ; the proof for $x<0$ is similar. $\mathcal{J}_{T,x}$ , as defined by (3.2), can be estimated as follows. By using Lemma 3.1, we have

\begin{equation*} \sup_{t\in\mathbb{R}}|\mathbb{Q}(\widetilde{G}_{T,x}\leq t) - \Phi(t)| \leq \frac{1}{\pi}\int_{-({|b|}/{32})T^{{1}/{2}}}^{({|b|}/{32})T^{{1}/{2}}} \bigg|\frac{\varphi_{\mathbb{Q},\widetilde{G}_{T,x}}(u) - \textrm{e}^{-{u^2}/{2}}}{u}\bigg|\,\textrm{d}{u} + \frac{768}{|b|\pi\sqrt{2\pi}\,}T^{-{1}/{2}}. \end{equation*}

Moreover, we have

\begin{align*} & \frac{1}{\pi}\int_{-({|b|}/{32})T^{{1}/{2}}}^{({|b|}/{32})T^{{1}/{2}}} \bigg|\frac{\varphi_{\mathbb{Q},\widetilde{G}_{T,x}}(u) - \textrm{e}^{-{u^2}/{2}}}{u}\bigg|\,\textrm{d}{u} \\[5pt] & \qquad \leq \frac{1}{\pi}\int_{-T^{{1}/{6}}}^{T^{{1}/{6}}} \bigg|\frac{\varphi_{\mathbb{Q},\widetilde{G}_{T,x}}(u) - \textrm{e}^{-{u^2}/{2}}}{u}\bigg|\,\textrm{d}{u} + \frac{1}{\pi}\int_{-({|b|}/{32})T^{{1}/{2}}}^{-T^{{1}/{6}}} \bigg|\frac{\varphi_{\mathbb{Q},\widetilde{G}_{T,x}}(u) - \textrm{e}^{-{u^2}/{2}}}{u}\bigg|\,\textrm{d}{u} \\[5pt] & \qquad\quad + \frac{1}{\pi}\int_{T^{{1}/{6}}}^{({|b|}/{32})T^{{1}/{2}}} \bigg|\frac{\varphi_{\mathbb{Q},\widetilde{G}_{T,x}}(u) - \textrm{e}^{-{u^2}/{2}}}{u}\bigg|\,\textrm{d}{u}. \end{align*}

From Proposition 2.1, it follows that, for $|u|\leq T^{{1}/{6}}$ ,

\begin{align*} \varphi_{\mathbb{Q},\widetilde G_{T,x}}({u}) & = \textrm{e}^{-\textrm{i}ux}\varphi_{\mathbb{Q},G_{T,x}}\bigg({u\sqrt{-\frac{2(b+1)}{T}}}\bigg) \\[5pt] & = \textrm{e}^{-{u^2}/{2}}\exp\{u\textrm{i}O(x^2T^{-{1}/{2}}) + u^2O(xT^{-{1}/{2}}) + O(u^3T^{-{1}/{2}})\}. \end{align*}

Together with Proposition 2.2, we have

\begin{align*} & \frac{1}{\pi}\int_{-({|b|}/{32})T^{{1}/{2}}}^{({|b|}/{32})T^{{1}/{2}}} \bigg|\frac{\varphi_{\mathbb{Q},\widetilde{G}_{T,x}}(u) - \textrm{e}^{-{u^2}/{2}}}{u}\bigg|\,\textrm{d}{u} \\[5pt] & \qquad \leq \frac{1}{\pi}\int_{-T^{{1}/{6}}}^{T^{{1}/{6}}}\textrm{e}^{-{u^2}/{2}} (O(x^2T^{-{1}/{2}}) + uO(xT^{-{1}/{2}}) + O(u^2T^{-{1}/{2}}))\,\textrm{d}{u} \\[5pt] & \qquad\quad + \frac{2}{\pi}\int_{T^{{1}/{6}}}^{({|b|}/{32})T^{{1}/{2}}} \frac{|\varphi_{\mathbb{Q},G_{T,x}}({u\sqrt{-{2(b+1)}/{T}}})| + \textrm{e}^{-{u^2}/{2}}}{u}\,\textrm{d}{u} \\[5pt] & \qquad = O((x^2 + 1)T^{-{1}/{2}}). \end{align*}

Therefore, we obtain

(3.3) \begin{equation} |\mathcal{J}_{T,x}| \leq \sup_{t\in\mathbb{R}}|\mathbb{Q}(\widetilde{G}_{T,x}\leq t) - \Phi(t)| = O((x^2 + 1)T^{-{1}/{2}}). \end{equation}

Applying (2.7), (2.8), (3.1), and (3.3),

\begin{align*} \mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b) > x\bigg) & = \mathbb{P}_b\bigg(G_{T,x}>x\sqrt{-\frac{T}{2(b+1)}}\bigg) \\[5pt] & = (1-\Phi(x))(1+O(x(x^2+1)T^{-{1}/{2}})). \end{align*}

Therefore, we get

\begin{align*} \bigg|\mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b) > x\bigg) - (1-\Phi(x))\bigg| & = (1-\Phi(x))O(x(x^2+1)T^{-{1}/{2}}) \\[5pt] & = \textrm{e}^{-{x^2}/{2}}O((x^2+1)T^{-{1}/{2}}), \end{align*}

which completes the proof of (1.4).

Note that

(3.4) \begin{equation} \frac{x}{\sqrt{2\pi}(1+x^2)}\exp\!\bigg\{{-}\frac{x^2}{2}\bigg\} \leq 1-\Phi(x) \leq \frac{1}{\sqrt{2\pi}x}\exp\!\bigg\{{-}\frac{x^2}{2}\bigg\}, \qquad x>0. \end{equation}

Together with (1.4), we can get (1.5), i.e.

\begin{equation*} \frac{\mathbb{P}_b(\sqrt{-{T}/({2(1+b)})}(\widehat b_T-b) \leq x)}{\Phi(x)} = \exp\!\bigg\{O(1)\frac{x^3}{\sqrt{T}\,}\bigg\}. \end{equation*}

Proof of Corollary 1.1. According to Theorem 1.1, for any $\rho>0$ ,

(3.5) \begin{equation} \sup_{x\leq\rho T^{{1}/{6}}}\bigg|\mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b) \leq x\bigg) - \Phi(x)\bigg| \leq CT^{-{1}/{2}}. \end{equation}

Moreover, using [Reference Jiang, Gao and Zhao14, Theorem 1.1] and (3.4), we have

(3.6) \begin{multline} \sup_{x\geq\rho T^{{1}/{6}}}\bigg|\mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b) \leq x\bigg) - \Phi(x)\bigg| \\[5pt] \leq \sup_{x\geq\rho T^{{1}/{6}}}\bigg|\mathbb{P}_b\bigg(\sqrt{-\frac{T}{2(1+b)}}(\widehat b_T-b) > x\bigg)\bigg| + \sup_{x\geq\rho T^{{1}/{6}}}(1-\Phi(x)) \leq CT^{-{1}/{2}}. \end{multline}

By (3.5) and (3.6), we can complete the proof of Corollary 1.1.

Proof of Corollary 1.2. Finally, Corollary 1.2 can be obtained immediately from (1.5) in Theorem 1.1.

Appendix A. Estimations of $\mathcal{R}_{\textbf{\textit{T,x}}}(\textbf{u})$ and $\mathcal{R}_{\textbf{\textit{T,x}}}(\textbf{0})$

In this appendix, we give precise estimates of $\mathcal{R}_{T,x}(u)$ and $\mathcal{R}_{T,x}(0)$ , which play crucial roles in the proof of Proposition 2.1.

Proposition A.1. For any constant $\rho>0$ , and $|u|\leq\rho T^{{1}/{6}}$ , $|x|\leq\rho T^{{1}/{6}}$ ,

(A.1) \begin{equation} \mathcal{R}_{T,x}(u) - \mathcal{R}_{T,x}(0) = O(T^{-{1}/{2}})u\textrm{i} + O(u^2T^{-1}) + O(u^3T^{-{3}/{2}}), \qquad \mathcal{R}_{T,x}(0) = O(xT^{-{1}/{2}}). \end{equation}

Proof. From (2.12) and (2.13), we have

(A.2) \begin{align} \varphi_{T,x}(u) & = O(x^2T^{-1}) + O(x^2T^{-{3}/{2}})u\textrm{i} + O(u^2T^{-1}) + O(u^3T^{-{3}/{2}}), \end{align}
(A.3) \begin{align} c_{T,x}(0) - b & = O(xT^{-{1}/{2}}), \end{align}
(A.4) \begin{align} c_{T,x}(u) - c_{T,x}(0) & = O(T^{-{1}/{2}})u\textrm{i} + O(u^2T^{-1}) + O(u^3T^{-{3}/{2}}). \end{align}

By Lemma 2.2, we have the following asymptotic results:

(A.5) \begin{align} \frac{\Gamma(\!-\!c_{T,x}(u)-({1}/{2})\varphi_{T,x}(u))}{\Gamma(\!-\!c_{T,x}(u))} & = \frac{-c_{T,x}(u)}{-c_{T,x}(u)-({1}/{2})\varphi_{T,x}(u)}\exp\{O(\varphi_{T,x}(u))\} \nonumber \\[5pt] & = 1 + O(\varphi_{T,x}(u)), \nonumber \\[5pt] \frac{\Gamma(\!-\!c_{T,x}(u) + {1}/{2})}{\Gamma(\!-\!c_{T,x}(u) - ({1}/{2})\varphi_{T,x}(u) + {1}/{2})} & = \frac{-c_{T,x}(u)-({1}/{2})\varphi_{T,x}(u)+{1}/{2}}{-c_{T,x}(u)+{1}/{2}} \exp\{O(\varphi_{T,x}(u))\} \nonumber \\[5pt] & = 1 + O(\varphi_{T,x}(u)), \nonumber \\[5pt] B\bigg(\!-\!c_{T,x}(u) - \frac{1}{2}\varphi_{T,x}(u),\frac{1}{2}\bigg) & = \frac{\Gamma(\!-\!c_{T,x}(u) - ({1}/{2})\varphi_{T,x}(u))\Gamma({1}/{2})} {\Gamma(\!-\!c_{T,x}(u) - ({1}/{2})\varphi_{T,x}(u) + {1}/{2})} \nonumber \\[5pt] & = B\bigg({-}c_{T,x}(u),\frac{1}{2}\bigg)(1 + O(\varphi_{T,x}(u))). \end{align}

Similarly, we also have

(A.6) \begin{equation} \frac{B(\!-\!c_{T,x}(u),{1}/{2})}{B(\!-\!c_{T,x}(0),{1}/{2})} = 1 + O(c_{T,x}(u) - c_{T,x}(0)), \qquad \frac{B(\!-\!c_{T,x}(0),{1}/{2})}{B(\!-\!b,{1}/{2})} = 1 + O(c_{T,x}(0) - b). \end{equation}

Recall the Jacobi Theta function [Reference Biane, Pitman and Yor6]:

\begin{align*} \Theta(z) = \sum_{\ell=-\infty}^{\infty}\textrm{e}^{-\pi\ell^2 z} = 1 + 2\sum_{\ell=1}^{\infty}\textrm{e}^{-\pi\ell^2 z}. \end{align*}

We have

\begin{equation*} \Theta(z) = \sum_{\ell=-\infty}^{\infty}\textrm{e}^{-\pi(\ell+1)^2 z} \leq \sum_{\ell=-\infty}^{\infty}\textrm{e}^{-\pi(\ell+{1}/{2})^2 z} \leq \sum_{\ell=-\infty}^{\infty}\textrm{e}^{-\pi\ell^2 z} = \Theta(z). \end{equation*}

Moreover, from the fact that

\begin{equation*} \sum_{\ell\geq0}\exp\!\bigg\{{-}\frac{1}{2}\pi^2\bigg(\ell+\frac{1}{2}\bigg)^2\bigg(\frac{s}{T(s+T)}\bigg)\bigg\} = \frac{1}{2}\Theta\bigg(\frac{\pi s}{2T(s+T)}\bigg), \end{equation*}

it follows that

\begin{align*} & \sum_{\ell\geq0}\int_{T}^{\infty}\exp\!\bigg\{{-}\frac{1}{2}\bigg[ (2n+2k+\gamma_1)^2(s-T)+\pi^2\bigg(\ell+\frac{1}{2}\bigg)^2\bigg(\frac{s-T}{Ts}\bigg)\bigg]\bigg\} \frac{\textrm{d}{s}}{\sqrt{Ts}\,} \\[5pt] & \qquad = \sum_{l\geq0}\int_{0}^{\infty}\exp\!\bigg\{{-}\frac{1}{2}\bigg[ (2n+2k+\gamma_1)^2s+\pi^2\bigg(\ell+\frac{1}{2}\bigg)^2\bigg(\frac{s}{T(s+T)}\bigg)\bigg]\bigg\} \frac{\textrm{d}{s}}{\sqrt{T(s+T)}\,} \\[5pt] & \qquad = \frac{1}{2}\int_{0}^{\infty}\exp\!\bigg\{{-}\frac{1}{2}(2n+2k+\gamma_1)^2s\bigg\} \Theta\bigg(\frac{\pi s}{2T(s+T)}\bigg)\frac{\textrm{d}{s}}{\sqrt{T(s+T)}\,}. \end{align*}

Recall the fact that $\Theta(x)=(1/\sqrt{x})\Theta (1/x)$ , which yields

\begin{multline*} \frac{1}{2}\int_{0}^{\infty}\exp\!\bigg\{{-}\frac{1}{2}(2n+2k+\gamma_1)^2s\bigg\} \Theta\bigg(\frac{\pi s}{2T(s+T)}\bigg)\frac{\textrm{d}{s}}{\sqrt{T(s+T)}\,} \\[5pt] = \frac{1}{\sqrt{2\pi}\,}\int_{0}^{\infty}\exp\!\bigg\{{-}\frac{1}{2}(2n+2k+\gamma_1)^2s\bigg\} \Theta\bigg(\frac{2T(s+T)}{\pi s}\bigg)\frac{\textrm{d}{s}}{\sqrt{s}\,}. \end{multline*}

By the asymptotic expansion of the Jacobi Theta function,

\begin{equation*} \Theta\bigg(\frac{2T(s+T)}{\pi s}\bigg) = 1 + 2\textrm{e}^{-{2T(s+T)}/{s}} + o(\textrm{e}^{-2T}) \leq 1 + 2\textrm{e}^{-2T} + o(\textrm{e}^{-2T}), \end{equation*}

we obtain

\begin{multline*} \sum_{l\geq0}\int_{T}^{\infty}\exp\!\bigg\{{-}\frac{1}{2}\bigg((2n+2k+\gamma_1)^2(s-T) + \pi^2\bigg(\ell+\frac{1}{2}\bigg)^2\bigg(\frac{s-T}{Ts}\bigg)\bigg)\bigg\}\frac{\textrm{d}{s}}{\sqrt{Ts}\,} \\[5pt] = \frac{1}{\sqrt{2\pi}\,}(1+O(\textrm{e}^{-2T}))\int_{0}^{\infty}\exp\!\bigg\{{-}\frac{1}{2}((2n+2k+\gamma_1)^2s)\bigg\} \frac{\textrm{d}{s}}{\sqrt{s}\,}. \end{multline*}

Consequently, we have

\begin{align*} & \frac{\textrm{e}^{\gamma_1^2{T}/{2}}}{\sqrt{T}}f_{T_1}*f_{C_{2n+\gamma_1}}\bigg(\frac{1}{T}\bigg) \\[5pt] & = \sum_{k,\ell\geq0}\frac{(\!-\!1)^k}{\sqrt{2\pi}}U_{k,n}\textrm{e}^{-({(2n+2k)^2}/{2})T-(2n+2k)\gamma_1T} \\[5pt] & \qquad \times \int_{T}^{\infty}\exp\!\bigg\{{-}\frac{1}{2}\bigg((2n+2k+\gamma_1)^2(s-T) + \pi^2\bigg(\ell+\frac{1}{2}\bigg)^2\bigg(\frac{s-T}{Ts}\bigg)\bigg)\bigg\} \frac{\textrm{d}{s}}{\sqrt{s}\,} \\[5pt] & = \frac{2^{2n+\gamma_1}}{\sqrt{2\pi}}\exp\{-2n^2T-2n\gamma_1T\} \sum_{k\geq0}\frac{(\!-\!1)^k\Gamma(2n+k+\gamma_1)}{k!\Gamma(2n+\gamma_1)}\textrm{e}^{-2k^2T-4nkT-2k\gamma_1T}(1+O(\textrm{e}^{-2T})). \end{align*}

Using Lemma 2.2,

\begin{equation*} \frac{\Gamma(2n+k+\gamma_1)}{\Gamma(2n+\gamma_1)} = \frac{2n+\gamma_1}{2n+k+\gamma_1}\exp\!\bigg\{-k\gamma + \sum_{i\geq1}\frac{k(2n+k+\gamma_1)}{i(i+2n+k+\gamma_1)} + {O\bigg(\bigg|\frac{k}{2n+k+\gamma_1}\bigg|^2\bigg)}\bigg\}, \end{equation*}

so we obtain

\begin{equation*} \sum_{k\geq0}\frac{(\!-\!1)^k\Gamma(2n+k+\gamma_1)}{k!\Gamma(2n+\gamma_1)}\textrm{e}^{-2k^2T-4nkT-2k\gamma_1T} = 1 + o(\textrm{e}^{-2T}), \end{equation*}

and the small $o(\cdot)$ term is hold uniformly in n; this further implies that

\begin{equation*} \frac{\textrm{e}^{\gamma_1^2{T}/{2}}}{\sqrt{T}}f_{T_1}*f_{C_{2n+\gamma_1}}\bigg(\frac{1}{T}\bigg) = \frac{2^{2n+\gamma_1}}{\sqrt{2\pi}}\exp\{-2n^2T-2n\gamma_1T\}(1 + O(\textrm{e}^{-2T})). \end{equation*}

Similarly, we have

(A.7) \begin{align} & \frac{\textrm{e}^{\gamma_1^2{T}/{2}}}{\sqrt{T}} \sum_{n=0}^{\infty}\frac{\Gamma(2n-c_{T,x}(u)+{1}/{2})}{4^nn!\Gamma(n-c_{T,x}(u))}f_{T_1}*f_{C_{2n+\gamma_1}} \bigg(\frac{1}{T}\bigg)B\bigg(n-c_{T,x}(u)-\frac{1}{2}\varphi_{T,x}(u),\frac{1}{2}\bigg) \nonumber \\[5pt] & = \frac{2^{\gamma_1}\Gamma(\!-\!c_{T,x}(u)+{1}/{2})}{\sqrt{2\pi}\Gamma(\!-\!c_{T,x}(u))} B\bigg({-}c_{T,x}(u)-\frac{1}{2}\varphi_{T,x}(u),\frac{1}{2}\bigg)(1 + O(\textrm{e}^{-2T})) \nonumber \\[5pt] & \quad \times \sum_{n=0}^{\infty}\frac{\Gamma(2n-c_{T,x}(u)+{1}/{2})\Gamma(\!-\!c_{T,x}(u)) B(n-c_{T,x}(u)-({1}/{2})\varphi_{T,x}(u),{1}/{2})}{n!\Gamma(\!-\!c_{T,x}(u)+{1}/{2}) \Gamma(n-c_{T,x}(u))B(\!-\!c_{T,x}(u)-{1}/{2})}\textrm{e}^{-2n^2T-2n\gamma_1T} \nonumber \\[5pt] & = \frac{2^{\gamma_1}\Gamma(\!-\!c_{T,x}(u)+{1}/{2})}{\sqrt{2\pi}\Gamma(\!-\!c_{T,x}(u))} B\bigg({-}c_{T,x}(u)-\frac{1}{2}\varphi_{T,x}(u),\frac{1}{2}\bigg)(1 + O(\textrm{e}^{-2T})). \end{align}

Applying (A.5) and (A.7), we get

(A.8) \begin{align} & \sqrt{2\pi}K_{\alpha_1}\frac{\textrm{e}^{\gamma_1^2{T}/{2}}}{\sqrt{T}} \sum_{n=0}^{\infty}\frac{\Gamma(2n-c_{T,x}(u)+{1}/{2})}{4^nn!\Gamma(n-c_{T,x}(u))}f_{T_1}*f_{C_{2n+\gamma_1}} \bigg(\frac{1}{T}\bigg)B\bigg(n-c_{T,x}(u)-\frac{1}{2}\varphi_{T,x}(u),\frac{1}{2}\bigg) \nonumber \\[5pt] & \qquad = B\bigg({-}c_{T,x}(u)-\frac{1}{2}\varphi_{T,x}(u),\frac{1}{2}\bigg)(1 + O(\textrm{e}^{-2T})) \nonumber \\[5pt] & \qquad = B\bigg({-}c_{T,x}(u),\frac{1}{2}\bigg)(1 + O(\varphi_{T,x}(u)))(1 + O(\textrm{e}^{-2T})). \end{align}

Combining this with (A.6) and (A.2)–(A.4), we have (A.1).

Acknowledgements

We would like to express our great gratitude to the two anonymous referees and the AE for their careful reading and insightful comments, which surely led to an improved presentation of this paper.

Funding information

Hui Jiang is supported by the Natural Science Foundation of Jiangsu Province of China (No. BK20231435) and Fundamental Research Funds for the Central Universities (No. NS2022069). Shaochen Wang is partially supported by Guangdong Basic and Applied Basic Research Foundation (No. 2023A1515012125).

Competing interests

There were no competing interests to declare which arose during the preparation or publication process of this article.

References

Ackerer, D., Filipović, D. and Pulido, S. (2018). The Jacobi stochastic volatility model. Finance Stoch. 22, 667700.10.1007/s00780-018-0364-8CrossRefGoogle Scholar
Ahdida, A. and Alfonsi, A. (2013). A mean-reverting SDE on correlation matrices. Stoch. Process. Appl. 129, 14721520.10.1016/j.spa.2012.12.008CrossRefGoogle Scholar
Ahlfors, L. V. (1979). Complex Analysis: An Introduction to the Theory of Analytic Functions of One Complex Variable, 3rd edn. McGraw-Hill, New York.Google Scholar
Bercu, B. and Richou, A. (2015). Large deviations for the Ornstein–Uhlenbeck process with shift. Adv. Appl. Prob. 47, 880901.10.1239/aap/1444308886CrossRefGoogle Scholar
Bernis, G. and Scotti, S. (2017). Alternative to beta coefficients in the context of diffusions. Quant. Finance 17, 275288.10.1080/14697688.2016.1188214CrossRefGoogle Scholar
Biane, P., Pitman, J. and Yor, M. (2001). Probability laws related to the Jacobi theta and Riemann zeta functions, and Brownian excursions. Bull. Amer. Math. Soc. 38, 435465.10.1090/S0273-0979-01-00912-0CrossRefGoogle Scholar
Delbane, F. and Shirakawa, H. (2002). An interest rate model with upper and lower bound. Asia-Pacific Financial Markets 9, 191209.10.1023/A:1024125430287CrossRefGoogle Scholar
Demni, N. and Zani, M. (2009). Large deviations for statistics of Jacobi process. Stoch. Process. Appl. 119, 518533.10.1016/j.spa.2008.02.015CrossRefGoogle Scholar
Ethier, S. N. and Kurtz, T. G. (2005) Markov Processes: Characterizations and Convergence, 2nd edn. John Wiley, New York.Google Scholar
Fan, X. Q., Ion, G., Liu, Q. S. and Shao, Q. M. (2019). Self-normalized Cramér type moderate deviations for martingales. Bernoulli 25, 27932823.10.3150/18-BEJ1071CrossRefGoogle Scholar
Fan, X. Q. and Shao, Q. M. (2023) Cramér’s moderate deviations for martingales with applications. Submitted.Google Scholar
Florens-Landais, D. and Pham, H. (1999). Large deviations in estimate of an Ornstein–Uhlenbeck model. J. Appl. Prob. 36, 6077.10.1239/jap/1032374229CrossRefGoogle Scholar
Gouriéroux, C. and Jasiak, J. (2006). Multivariate Jacobi process with application to smooth transitions. J. Econometrics 131, 475505.10.1016/j.jeconom.2005.01.014CrossRefGoogle Scholar
Jiang, H., Gao, F. Q. and Zhao, S. J. (2009). Moderate deviations for statistics of Jacobi process. Chinese Ann. Math. Ser. A 30, 479490.Google Scholar
Jiang, H. and Zhou, J. Y. (2023). An exponential nonuniform Berry–Esseen bound of the fractional Ornstein–Uhlenbeck process. J. Theoret. Prob. 36, 10371058.10.1007/s10959-022-01194-wCrossRefGoogle Scholar
Karlin, S. and Taylor, H. M. (1981). A Second Course in Stochastic Processes, 1st edn. Academic Press, New York.Google Scholar
Kutoyants, Y. A. (2004). Statistical Inference for Ergodic Diffusion Process, 1st edn. Springer, London.10.1007/978-1-4471-3866-2CrossRefGoogle Scholar
Petrov, V. (1975). Sums of Independent Random Variables, 1st edn. Springer, New York.Google Scholar
Wang, Q. Y. and Jing, B. Y. (1999). An exponential nonuniform Berry–Esseen bound for self-normalized sums. Ann. Prob. 27, 20682088.Google Scholar
Zang, Q. P. and Zhang, L. X. (2016). A general lower bound of parameter estimation for reflected Ornstein–Uhlenbeck processes. J. Appl. Prob., 53, 2232.10.1017/jpr.2015.5CrossRefGoogle Scholar