Confidence interval for correlation estimator between latent processes
Extracted source text ยท arXiv 1710.06683
\documentclass[a4paper]{article}
\usepackage{amssymb}
\usepackage{amsthm}
\usepackage{amsmath}
\usepackage{enumerate}%\begin{enumerate}[{(}i{)}]
\usepackage{bm}
% \usepackage{bbm} %\mathbbm{ABCabc\nabla}
% \usepackage[bbgreekl]{mathbbol} %\mathbb{abcABC\nabla\bbalpha\bbbeta\Delta }
\newtheorem{theorem*}{Theorem}[section]
\newtheorem{ass}[theorem*]{Assumption}
\newtheorem{note*}[theorem*]{Note}
\newtheorem{lemma*}[theorem*]{Lemma}
\newtheorem{definition*}[theorem*]{Definition}
\newtheorem{proposition*}[theorem*]{Proposition}
\newtheorem{corollary*}[theorem*]{Corollary} %{corollary}
\newtheorem{remark*}[theorem*]{Remark}
\newtheorem{example*}[theorem*]{Example}
\numberwithin{equation}{section}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{document}
\title{
Confidence interval for correlation estimator between latent processes
%Confidence interval estimation of correlation between latent %intensity
%processes
%for aeXiv.
%\footnote{
%This work was in part supported by
%Japan Society for the Promotion of Science Grants-in-Aid for Scientific Research No. 24340015 (Scientific Research),
%No. 24650148 (Challenging Exploratory Research);
%and by a Cooperative Research Program of the Institute of Statistical Mathematics.
%}
%\protect\footnote
%\thanks{
%This work was in part supported by
%Japan Society for the Promotion of Science Grants-in-Aid for Scientific Research Nos. 24340015
%%and 24300107 %Uchida's
%(Scientific Research),
%Nos. 24650148 and 26540011 (Challenging Exploratory Research);
%%the Global COE program ``The Research and Training Center for New Development in Mathematics''
%%of the Graduate School of Mathematical Sciences, University of Tokyo;
%CREST Japan Science and Technology Agency;
%NS Solutions Corporation;
%and by a Cooperative Research Program of the Institute of Statistical Mathematics.
%%{\colorred The authors thank the referee for valuable comments.}
%}
}
\author{
Akitoshi Kimura %\inst{1} and Nakahiro Yoshida\inst{2}
%Name of First Author and Name of Second Author
\thanks{Graduate School of Mathematical Sciences, University of Tokyo, 3-8-1 Komaba, Meguro-ku, Tokyo 153-8914, Japan, %\email{kimura@ms.u-tokyo.ac.jp}
Japan Science and Technology Agency, CREST}
}
% Use \authorrunning{Short Title} for an abbreviated version of
% your contribution title if the original one is too long
%\institute{Akitoshi Kimura \at \inst{1}Graduate School of Mathematical Sciences, University of Tokyo, 3-8-1 Komaba, Meguro-ku, Tokyo 153-8914, Japan, \email{kimura@ms.u-tokyo.ac.jp}
%\at \inst{1}Japan Science and Technology Agency, CREST
%\and Nakahiro Yoshida \at \inst{2}Graduate School of Mathematical Sciences, University of Tokyo, 3-8-1 Komaba, Meguro-ku, Tokyo 153-8914, Japan, \email{narahiro@ms.u-tokyo.ac.jp}
%\at \inst{2} Japan Science and Technology Agency, CREST }
%
% Use the package "url.sty" to avoid
% problems with special characters
% used in your e-mail or web address
%
\maketitle
\begin{abstract}
Kimura and Yoshida \cite{Kimura2016} treated a model
in which
the finite variation part of a two-dimensional semimartingale is expressed by time-integration of latent %It\^o
processes.
%
%including doubly stochastic Poisson processes whose intensity processes are correlated It\^o processes.
%In order to estimate the correlation between intensity processes of doubly stochastic Poisson processes,
%\cite{Kimura2016} introduced the more generalized model:
%In the model,
%finite variation part of two-dimensional semi-martingale is expressed by time-integration of latent It\^o processes.
%\cite{Kimura2016}
They proposed a correlation estimator between the latent processes and proved its consistency and asymptotic mixed normality.
In this paper, we discuss the confidence interval of the correlation estimator to detect the correlation. %between latent processes.
We propose two types of estimators for asymptotic variance of the correlation estimator and prove their consistency in a high frequency setting.
Our model includes %including
doubly stochastic Poisson processes whose intensity processes are correlated It\^o processes.
We compare %their performance
our estimators based on %in
the simulation of the %estimation for correlation between intensity processes of
doubly stochastic Poisson processes.
%
%In order to detect the correlation between latent processes, we discuss the confidence interval of the correlation estimator.
%We propose two types of asymptotic variance estimators and prove the consistency in a high frequency setting.
%We compare these in the simulation of the estimation of correlation between latent processes of doubly stochastic point processes.
\end{abstract}
\section{Introduction}
%
%\begin{itemize}
%\item
%WHY is the model (counting process) introduced? [reference]...
%\item
%WHY do we treat the quantity (correlation) ? [reference]...
%\cite{zhou1996high}
%\end{itemize}
%
Integrated correlation is an important index in high frequency financial data analysis.
%The phenomenon that the sample correlation between the returns of two different stocks decreases as the sampling frequency of data increases is well-known as Epps effect.
Epps \cite{epps1979comovements} pointed out that the sample correlation between the returns of two different stocks decreases as the sampling frequency of data increases.
It is considered that non-synchronicity and market microstructure of trading cause this phenomenon.
%In recent years,
Non-synchronous covariance estimation schemes %, for example,
have been developed:
Fourier analytic approach
(Malliavin and Mancino \cite{malliavin2002fourier},
Malliavin et al. \cite{Malliavin2007}) and
the cumulative covariance estimator
(Hayashi and Yoshida \cite{%HayashiYoshida2005,HayashiYoshida2008,HayashiYoshida2011
hayashi2005covariance,hayashi2008asymptotic,hayashi2011nonsynchronous},
Mykland \cite{mykland2012gaussian}). %;
%and denoising techniques, for example,
%have been developed in recent years:
%Of course, there
The
%On the orher hand, the
market microstructure is %often expressed
modeled
as the noise added to the latent price process.
This modeling was successful and
denoising techniques have also been developed:
sub-sampling
(Zhang et al. \cite{%zhang2005tale
zhang2012tale},
Zhang \cite{zhang2006efficient}),
pre-averaging
(Podolskij and Vetter \cite{%podolskij2009estimation
podolskij2007estimation},
Jacod et al. \cite{jacod2009microstructure}),
%kernel method (),
and others (Zhou \cite{zhou1996high}).
%
There are many studies that treat both non-synchoronicity and market microstructure noise:
Malliavin and Mancino \cite{malliavin2009fourier},
Mancino and Sanfelici \cite{%RePEc:eee:csdana:v:52:y:2008:i:6:p:2966-2989,
mancino2011estimating},
Park and Linton \cite{park2012estimating},
Voev and Lunde \cite{voev2007integrated},
Griffin and Oomen \cite{griffin2011covariance},
Christensen et al. \cite{christensen2010pre, %},
%Christensen et al. \cite{
christensen2013covariation}, %20170826 add % how do I explain?
Koike \cite{koike2013estimation, koike2014limit,koike2014estimator},
A{\"\i}t-Sahalia et al. \cite{ait2010high}, %20170826 add
Barndorff-Nielsen et al. \cite{barndorff2011multivariate}, %20170826 add
Bibinger \cite{bibinger2011efficient,bibinger2012estimator}.
%
%
%% Fourier approoach
%Malliavin and Mancino \cite{%MalliavinMancino2002
%malliavin2002fourier,malliavin2009fourier},
%% add 20170809 %
%Malliavin et al.
%\cite{Malliavin2007},
%%
%%Cl\'ement and Gloter
%%\cite{CLEMENT20111097},
%%
%Mancino and Sanfelici
%\cite{%RePEc:eee:csdana:v:52:y:2008:i:6:p:2966-2989,
%mancino2011estimating},
%Park and Linton
%\cite{park2012estimating},
%% add 20170809 %
%% hayashi-yoshida approach
%Hayashi and Yoshida \cite{%HayashiYoshida2005,HayashiYoshida2008,HayashiYoshida2011
%hayashi2005covariance,hayashi2008asymptotic,hayashi2011nonsynchronous},
%Voev and Lunde \cite{voev2007integrated},
%Griffin and Oomen \cite{griffin2011covariance},
%Mykland \cite{mykland2012gaussian},
%%
%% Microstructure
%Zhou \cite{zhou1996high},
%Zhang et al. \cite{%zhang2005tale
%zhang2012tale},
%Zhang \cite{zhang2006efficient},
%%Podolskij and Better \cite{podolskij2009estimation},
%Podolskij and Vetter \cite{%podolskij2009estimation
%podolskij2007estimation},
%Jacod et al. \cite{jacod2009microstructure},
%%
%% Nonsynchronous + microstructure
%A{\"\i}t-Sahalia et al. \cite{ait2010high}, %20170826 add
%Barndorff-Nielsen et al. \cite{barndorff2011multivariate}, %20170826 add
%Christensen et al. \cite{christensen2010pre},
%Christensen et al. \cite{christensen2013covariation}, %20170826 add % how do I explain?
%Bibinger \cite{bibinger2011efficient,bibinger2012estimator},
%Ogihara and Yoshida \cite{ogihara2012quasi},
%Koike \cite{koike2013estimation, koike2014limit,koike2014estimator},
%Ogihara \cite{%ogihara2013local
%ogihara2015local,ogihara2014parametric}.
%
%
%As the
%Various causality of market microstructure are proposed: %,
%for example,
%
%%
%%The
%By the way, the lead-lag
%%Lead-lag
%phenomenon is also an issue related to the
%non-synchronicity: %,
%%see
%%e.g.,
%de Jong and Nijman \cite{%Jong1997
%de1997high},
%Hoffmann et al. \cite{%HoffmannRosenbaumYoshida2013
%hoffmann2013estimation} and Abergel and Huth \cite{abergel2012high}.
%%
% Recently, along with the developments in measurement and storage technologies, the timing of sampling is becoming more and more precise.
%There is no Brownian motion driving the system, differently from the standard framework, since the central limit theorem is not effective at this level of description.
%
%In ultra high frequency sampling,
%there is no Brownian motion driving the asset prices, %system,
%differently from the standard mathematical finance framework,
%since the central limit theorem does not work. %is not effective at this level of description.
%In order to model
%non-synchronicity
%%non-synchronisity
%and microstructure as well as lead-lags,
%modeling of ultra high frequency phenomena %is going toward use of
%with jump processes
%modeling limit order book (LOB) emerge:
%
The followings are proposed as the cause of market microstructure:
bid-ask spread
(Roll \cite{10.2307/2327617}),
discretization error
(Gottlieb and Kalay \cite{10.2307/2328052}),
and
asymmetric information
(Glosten and Milgrom \cite{GLOSTEN198571}). %,
%are proposed.
On the other hand,
the relationship between the additive noise modeling and the %causality of
market microstructure is not so clearly explained.
%at least for develiping theory.
%though the reality of the %noise
%modeling is not so clearly explained.
%In order to deal with
In ultra high frequency sampling, recently,
the market microstructure is modeled as dynamics of the limit order book (LOB) rather than the noise.
%In order to deal with non-synchronicity and market microstructure, %and lead-lags
Many studies adopt approaches modeling LOB with %Point processes:
Poisson processes
(Cont et al. \cite{cont2010stochastic}, %and
Abergel and Jedidi \cite{abergel2013mathematical},
Muni Toke
\cite{doi:10.1080/14697688.2014.963654, 2015arXiv150203871M}, %, 2010arXiv1003.3796M}
Smith et al. \cite{smith2003statistical}, %.
Muni Toke and Yoshida
\cite{2016arXiv160203944M}),
Hawkes processes
(Hewlett \cite{hewlett2006clustering}, % on the clustered arrivals of buy and sell trades using Hawkes processes,
Large \cite{large2007measuring}, % on extension by using a fine description of orders,
Bowsher \cite{bowsher2007modelling}, % on a generalized Hawkes model, and {\blue in}
Bacry et al. \cite{bacry2013modelling}, % on a price model.
Abergel and Jedidi \cite{%abergel2013mathematical,
abergel2015long}, %and also
Muni Toke and Pomponio \cite{muni2011modelling}, %gave a model of trades-through in a limited order book using Hawkes processes.
Muni Toke
\cite{%doi:10.1080/14697688.2014.963654, 2015arXiv150203871M,
2010arXiv1003.3796M},
Clinet and Yoshida
\cite{CLINET20171800},
Ogihara and Yoshida \cite{%OgiharaYoshida2015pprm
ogihara2015quasi}), and
doubly stochastic Poisson processes
(Abergel et al. \cite{2017arXiv170501446A},
Guilbaud and Pham \cite{2011arXiv1106.5040G},
Chertok et al. \cite{Chertok2016},
Korolev et al. \cite{2014arXiv1410.1900K}).
These approaches can also treat the non-synchronicity of trading.
%
%
%
%Hewlett \cite{hewlett2006clustering}, % on the clustered arrivals of buy and sell trades using Hawkes processes,
%Large \cite{large2007measuring}, % on extension by using a fine description of orders,
%Bowsher \cite{bowsher2007modelling}, % on a generalized Hawkes model, and {\blue in}
%Bacry et al. \cite{bacry2013modelling}, % on a price model.
%%Chen and Hall \cite{chen2013inference} investigated the maximum likelihood estimator
%%of a non-stationary self-exciting point process when the intensity goes up.
%%
%Cont et al. \cite{cont2010stochastic}, %and
%Abergel and Jedidi \cite{abergel2013mathematical, abergel2015long}, %and also
%Smith et al. \cite{smith2003statistical}, %.
%%Abergel and Jedidi \cite{abergel2013mathematical,abergel2015long} presented an order book model by a multivariate point process
%%and proved ergodicity of the system in infinite time horizon by using the drift condition for the Markov chain.
%Muni Toke and Pomponio \cite{muni2011modelling}, %gave a model of trades-through in a limited order book using Hawkes processes.
%Muni Toke
%\cite{doi:10.1080/14697688.2014.963654, 2015arXiv150203871M, 2010arXiv1003.3796M}
%Muni Toke and Yoshida
%\cite{2016arXiv160203944M},
%%
%Clinet and Yoshida
%\cite{CLINET20171800},
%%Multivariate point process; Ergodicity; Hawkes process; Inferential statistics; Quasi Likelihood Analysis; Limit Order Book
%%
%%Zheng et al.
%%\cite{doi:10.1137/130912980},
%%
%and
%Ogihara and Yoshida \cite{%OgiharaYoshida2015pprm
%ogihara2015quasi}.
%
%The latest financial statistics is directed toward analysis of ultra high frequency phenomena by modeling in a more and more precise time-scale.
%There is no Brownian motion driving the system, differently from the standard framework, since the central limit theorem is not effective at this level of description.
%Recently point processes are attracting attention in modeling limit order book (LOB).
%
%Jump processes have been applied to analysis of financial data.
In this stream, Kimura and Yoshida \cite{Kimura2016} focused on the integrated correlation between intensity processes of doubly stochastic Poisson processes.
%In this paper, we discuss the following model introduced in \cite{Kimura2016}. %and including doubly stochastic Poisson processes.
They introduced the following model to treat it in a generalized form.
This paper is also based on the model.
Now, we consider a stochastic basis
$\mathcal{B} = (\Omega, \mathcal{F}, \mathbb{F}, P)$,
$\mathbb{F} = (\mathcal{F}_{t})_{t \in [0,T]}$.
On $\mathcal{B}$, let
$\mathbb{X} = (X^1, X^2)$ be an $\mathbb{R}^2$-valued It\^o process given by
\begin{align} \label{Xint}
\mathbb{X}_{t} = \mathbb{X}_{0} + \int_{0}^{t} \mathbb{X}^{0}_{s} \ ds + \int_{0}^{t} \mathbb{X}^{1}_{s} \ dw_{s}
\ \ (t \in [0,T]),
\end{align}
where
$w$ is an $r$-dimensional $\mathcal{F}$-Wiener process,
$\mathbb{X}_{0}$ is an $\mathbb{F}_{0}$-measurable random variable,
$\mathbb{X}^{0}$ is a two-dimensional $\mathbb{F}$-adapted process, and
$\mathbb{X}^{1}$ is an $\mathbb{R}^{2} \otimes \mathbb{R}^{r}$-valued $\mathbb{F}$-adapted process
satisfies condition [A] mentioned later.
Let
$a_n$ be a positive number depending on $n$.
On $\mathcal{B}$, consider a two-dimensional measurable process
$\mathbb{Y}^n = (Y^{n, 1}, Y^{n, 2})$ having a decomposition
\begin{align*}
\mathbb{Y}^{n}_{t} = \mathbb{Y}^{n}_{0} + \int_{0}^{t} a_n \mathbb{X}_{s} \ ds + \mathbb{M}^{n}_{t}
\ \ (t \in [0,T]),
\end{align*}
where $\mathbb{M}^{n}$ is a two-dimensional measurable process with $\mathbb{M}^{n}_{0} = 0$
satisfies condition %[B$^{\flat}$],
[B], [B'], or [B$^{\sharp^{\sharp}}$] mentioned later.
In this model,
$\mathbb{Y}$ can be a counting process with intensity process $a_n \mathbb{X}$
where $\mathbb{X}$ is $\mathbb{R}_+^2$-valued,
i.e. $\mathbb{Y}$ is a two-dimensional doubly stochastic Poisson process.
This counting process models the high frequency counting data of the orders or transactions in the active market, for example.
Let $I_j = [t_{j-1}, t_{j})$ for a sampling design $\Pi = (t_j)_{j=0, \dots , b_n}$
with $0 = t_{0} < t_{1} < \dots < t_{b_n} =T$
and $h_{j} = t_{j} - t_{j-1}$.
Kimura and Yoshida \cite{Kimura2016} introduced the
%The
estimator of the covariance $\langle X^1, X^2 \rangle$ from the sampled data of $\mathbb{Y}$ %is introduced
as follows.
% Write $\Delta_{j} V = V_{t_{j}} - V_{t_{j-1}}$ for a process $V$.
For $\alpha, \beta = 1, 2$,
\begin{align*}
S_n^{\alpha \beta}
&= \sum_{k=2}^{b_n}
\Big(\frac{Y^{n, \alpha}_{t_{k}} - Y^{n, \alpha}_{t_{k-1}}}{a_n h_{k}} - \frac{Y^{n, \alpha}_{t_{k-1}} - Y^{n, \alpha}_{t_{k-2}}}{a_n h_{k-1}} \Big)
\Big(\frac{Y^{n, \beta}_{t_{k}} - Y^{n, \beta}_{t_{k-1}}}{a_n h_{k}} - \frac{Y^{n, \beta}_{t_{k-1}} - Y^{n, \beta}_{t_{k-2}}}{a_n h_{k-1}} \Big), % \\
%&= \sum_{k=2}^{b_n} (\tilde{\mathbb{Y}}^n_{k} - \tilde{\mathbb{Y}}^n_{k-1})^{\otimes(\alpha, \beta)}.
%= \sum_{j=2}^{b_n} (\frac{\Delta_{j} \mathbb{Y}^n}{a_n h_{j}} - \frac{\Delta_{j-1} \mathbb{Y}^n}{a_n h_{j-1}})^{\otimes(\alpha, \beta)},
%= \frac{1}{(a_n \delta_n)^2} \sum_{j=2}^{b_n} (\Delta_{j} \mathbb{Y}^n - \Delta_{j-1} \mathbb{Y}^n)^{\otimes(\alpha, \beta)},
\end{align*}
% where
% $A^{\otimes(i, j)}$ is an $(i, j)$-element of $A^{\otimes}$,
% $A^{\otimes} = A \otimes A = AA^{\star}$, and
% $^\star$ denotes transpose.
Here, $S_n = (S_n^{1 2}, S_n^{1 1}, S_n^{2 2})^{\star}$ is the (co)variance estimator.
$S_n$ depends on the scaling parameter $a_n$ which is not derived from the data of $\mathbb{Y}$.
Therefore, treat the correlation estimator $C_n^{1 2} = S_n^{1 2}/\sqrt{S_n^{1 1} S_n^{2 2}}$
% \begin{eqnarray*}
% C_n^{1 2} = \frac{S_n^{1 2}}{\sqrt{S_n^{1 1} S_n^{2 2}}}.
% \end{eqnarray*}
which does not depend on the scaling parameter $a_n$.
% Here, let
Let
\begin{align*}
U^{\alpha \beta} = \frac{2}{3} \langle X^{\alpha,c}, X^{\beta,c} \rangle_T
\>=\>\frac{2}{3}\int_0^T X^{\alpha1}_{t}\cdot X^{\beta1}_{t}dt
\quad(\alpha,\beta=1,2)
\end{align*}
and
%$U = (U^{\alpha \beta}, U^{\alpha \alpha}, U^{\beta \beta})^\star$.
$U = (U^{12}, U^{11}, U^{22})^\star$,
where $X^{\alpha,c}$ is the continuous part of $X^{\alpha}$
and $X^{\alpha1}_{t}$ is the $\alpha$-th row of $\mathbb{X}_t^{1}$.
%
%
%{\colorb !!!!!!! Original theorems were not correctly written. Correct them later. :
%
% Let $x\tilde{\otimes}y=((x_iy_j+x_jy_i)/2)\in\mathbb{R}^{\sf r}\otimes\mathbb{R}^{\sf r}$
% for $x=(x_i)$, $y=(y_i)\in\mathbb{R}^{\sf r}$, and
% let $x^{\tilde{\otimes}(\alpha,\beta)}=x^\alpha_\cdot\tilde{\otimes}x^\beta_\cdot$ for $x=(x^\alpha_i)\in\mathbb{R}^2\otimes\mathbb{R}^{\sf r}$.
% We write $x\cdot y=\sum_{i=1}^{\sf r}x_iy_i$ for $x=(x_i)$, $y=(y_i)\in\mathbb{R}^{\sf r}$,
% and $x\cdot y = \sum_{i,j=1}^{\sf r}x_{i,j}y_{i,j}$ for
% $x=(x_{i,j})$, $y=(y_{i,j})\in\mathbb{R}^{\sf r}\otimes \mathbb{R}^{\sf r}$.
Let
% For $\alpha_1, \beta_1, \alpha_2, \beta_2 = 1, 2$, let
\begin{align*}
\gamma^{(\alpha_1,\beta_1),(\alpha_2,\beta_2)}
&= \int_0^T \sum_{i,j=1}^{r}
\frac{X_{i,s}^{\alpha_1 1} X_{j,s}^{\beta_1 1} + X_{j,s}^{\alpha_1 1} X_{i,s}^{\beta_1 1}}{2}
\frac{X_{i,s}^{\alpha_2 1} X_{j,s}^{\beta_2 1} + X_{j,s}^{\alpha_2 1} X_{i,s}^{\beta_2 1}}{2} \ ds \\
&= \int_0^T (\mathbb{X}_s^{1})^{\tilde{\otimes} (\alpha_1,\beta_1)}
\cdot (\mathbb{X}_s^{1})^{\tilde{\otimes} (\alpha_2,\beta_2)} \ ds.
%\gamma ^{p q} = \int_0^1 (\mathbb{X}_s^{1})^{\tilde{\otimes} p} \cdot
%(\mathbb{X}_s^{1})^{\tilde{\otimes}q} \ ds,
\end{align*}
and $\Gamma = (\gamma ^{p q})_{p,q = (1, 2), (1,1), (2,2)}$,
where $X_{i,s}^{\alpha 1}$ is the $(\alpha, i)$-element of $\mathbb{X}_s^{1}$.
% where $x\tilde{\otimes}y=((x_iy_j+x_jy_i)/2)\in\mathbb{R}^{\sf r}\otimes\mathbb{R}^{\sf r}$
% for $x=(x_i)$, $y=(y_i)\in\mathbb{R}^{\sf r}$,
% $x^{\tilde{\otimes}(\alpha,\beta)}=x^\alpha_\cdot\tilde{\otimes}x^\beta_\cdot$ for $x=(x^\alpha_i)\in\mathbb{R}^2\otimes\mathbb{R}^{\sf r}$,
% $x\cdot y=\sum_{i=1}^{\sf r}x_iy_i$ for $x=(x_i)$, $y=(y_i)\in\mathbb{R}^{\sf r}$,
% and $x\cdot y = \sum_{i,j=1}^{\sf r}x_{i,j}y_{i,j}$ for
% $x=(x_{i,j})$, $y=(y_{i,j})\in\mathbb{R}^{\sf r}\otimes \mathbb{R}^{\sf r}$.
Under [A] and any one of [B], [B'], and [B$^{\sharp^{\sharp}}$], %some assumption,
Kimura and Yoshida \cite{Kimura2016} proved the the asymptotic mixed normality of the covariance estimator and the correlation estimator.
\begin{align*}
\Big(\frac{T}{b_n} \Big)^{-1/2} (S_n - U) \to^{d_s} \Gamma^{1/2} \zeta
\end{align*}
as $n \to \infty$,
where $\zeta$ is an $\mathbb{R}^3$-valued standard normal variable independent of $\mathcal{F}$.
and $d_s$ means $\mathcal{F}$-stable convergence.
Let $R = U^{12} / \sqrt{U^{11} U^{22}}$
and suppose that $U^{11} U^{22} \neq 0$ a.s.
\begin{align*}
\Big(\frac{T}{b_n} \Big)^{-1/2} (C_n^{12} - R) \to^{d_s} \Xi^{1/2} \zeta
\end{align*}
as $n \to \infty$, where
\begin{align*}
\Xi^{1/2} :=
\Big(\frac{1}{\sqrt{U^{11} U^{22}}}, \frac{-U^{12}}{2 \sqrt{(U^{11})^3 U^{22}}}, \frac{-U^{12}}{2 \sqrt{U^{11} (U^{22})^3}} \Big) \Gamma^{1/2}.
\end{align*}
$\Xi = (\Xi^{1/2}) (\Xi^{1/2})^{\star}$ is the asymptotic variance of the correlation estimator.
In this paper, we introduce two types of estimators for $\Gamma$ and prove their consistency
in order to obtain the consistent estimator for $\Xi$.
%Then we get the consistent asymptotic variance estimator.
%Hereafter,
Therefore, we can detect the correlation between latent processes based on hypothesis testing.
%This result can be applied to lead-lag detection.
This paper is organized as follows.
In Section \ref{assumption}, we list the assumptions. %confirm the assumptions in \cite{Kimura2016}.
In Section \ref{result}, we introduce the %asymptotic variance
estimators for $\Gamma$ and $\Xi$, and state the main result: their consistency. %prove the consistency.
In Section \ref{proof}, the results in Section \ref{result} are proved.
In Section \ref{simulation}, an example and simulation studies are given.
\section{Assumptions} \label{assumption}
For simplicity, we assume
$t_{j} = j (T/b_n)$, $h_{j} = T/b_n =: \delta_n$.
We write $\Delta_{j} V = V_{t_{j}} - V_{t_{j-1}}$ for a process $V$.
Consider the following conditions.
\begin{itemize}
\item[\bf{[A]}]
Process $\mathbb{X}$ admits the representation (\ref{Xint})
for an $\mathbb{R}^2$-valued $\mathcal{F}_0$-measurable random variable $\mathbb{X}_0$
and coefficients $\mathbb{X}^{\kappa}$ ($\kappa=0,1$) such that
$\mathbb{X}^{0}$ is a c\'adl\'ag $\mathbb{F}$-adapted process and that
$\mathbb{X}^{1}$ has a representation
\begin{align*}
\mathbb{X}^{1}_{t}
= \mathbb{X}^{1}_{0} + \int_{0}^{t} \mathbb{X}^{10}_{s} \ ds + \int_{0}^{t} \sum_{\kappa' = 1}^{r'} \mathbb{X}^{1\kappa'}_{s} \ d\tilde{w}^{\kappa'}_s
\ \ (t \in [0,T]),
\end{align*}
where
$\mathbb{X}^{1}_{0}$ is an $\mathbb{R}^{2} \otimes \mathbb{R}^{r}$-valued $\mathcal{F}_{0}$-measurable random variable,
$\tilde{w} = (\tilde{w}^1,...,\tilde{w}^{r'})$ is an $r'$-dimensional $\mathbb{F}$-Wiener process (not necessary independent of $w$),
and
$\mathbb{X}^{1 \kappa'}_{s}$ $(\kappa' = 0, 1, ..., r')$ are $\mathbb{R}^{2} \otimes \mathbb{R}^{r}$-valued c\'adl\'ag $\mathbb{F}$-adapted processes.
%\item[\bf{
%[B$^\flat$]
%}]
%\begin{itemize}
%\item[
%(i)
%]
%$\lim_{n \to \infty} b_n^{2} / a_n = 0$.
%
%\item[
%(ii)
%]
%$\sum_{j=1}^{b_n} |\Delta_{j} \mathbb{M}^n|^2 = O_p(a_n)$ as $n \to \infty$.
%\end{itemize}
%
\item[\bf{
[B]
}]
$\mathbb{M}^n = (M^{n, \alpha})_{\alpha=1, 2}$ is a two-dimensional $\mathbb{F}$-local martingale
with $\mathbb{M}^n_{0} = 0$ and such that
\begin{itemize}
\item[
(i)
]
$\lim_{n \to \infty} b_n^{5/2} / a_n = 0$.
\item[
(ii)
]
$\sum_{j=1}^{b_n} |\Delta_{j} \mathbb{M}^n|^2 = O_p(a_n)$ as $n \to \infty$,
$\sup_{t \in [0,1]} |\Delta \mathbb{M}^n| \leq c a_n^{1/2}$ for a constant c independent of n.
\item[
(iii)
]
The absolutely continuous (w.r.t. the Lebesgue measure a.s.) mapping
$[0, T] \ni t \mapsto \langle \mathbb{M}^n , w \rangle_{t} \in \mathbb{R}^{2} \otimes \mathbb{R}^{r}$
satisfies
$\sup_{t \in [0, T]} |d \langle \mathbb{M}^n , w \rangle_{t} / dt| = O_p(b_n)$ as $n \to \infty$.
\end{itemize}
Here,
$\langle \mathbb{M}^n , w \rangle_{t}$ is the $2 \times r$ matrix of angle brackets $\langle M^{n, \alpha} , w^{k} \rangle_{t}$ for $\mathbb{M}^n = (M^{n, \alpha})_{\alpha = 1, 2}$ and $w = (w^k)_{k=1,...,r}$.
\item[\bf{
[B$'$]
}]
$\mathbb{M}^n$ is a two-dimensional $\mathbb{F}$-local martingale
with $\mathbb{M}^n_{0} = 0$, satisfies [B] (i), (iii), and
\begin{itemize}
\item[
(ii$'$)
]
$E[\sum_{j=1}^{b_n} |\Delta_{j} \mathbb{M}^n|^2] = O(a_n)$ as $n \to \infty$.
\end{itemize}
\item[\bf{
[B$^{\sharp^{\sharp}}$]
}]
$\mathbb{M}^n$ is a two-dimensional $\mathbb{F}$-local martingale
with $\mathbb{M}^n_{0} = 0$, satisfies [B] (ii), (iii), and
\begin{itemize}
\item[
(i$^{\sharp^{\sharp}}$)
]
$\lim_{n \to \infty} b_n^{3} / a_n = 0$.
%
%\item[
%(ii)
%]
%$\sum_{j=1}^{b_n} |\Delta_{j} \mathbb{M}^n|^2 = O_p(a_n)$ as $n \to \infty$
\end{itemize}
\item[\bf{
[C]
}]
$E[\sum_{j=1}^{b_n} |\Delta_{j} \mathbb{M}^n|^4] = O(a_n^2 b_n^{-1})$ as $n \to \infty$
\end{itemize}
\begin{remark*}
[A], [B], and [B$'$] are the same condition in \cite{Kimura2016}.
[B$^{\sharp^{\sharp}}$] is a bit stronger condition than [B$^{\sharp}$] in \cite{Kimura2016}.
\end{remark*}
\begin{remark*}
The doubly stochastic Poisson process model satisfies [C].
\end{remark*}
\section{Results} \label{result}
We need the estimator of $\Gamma$ to %get
obtain the asymptotic variance estimator.
Here, we give two types of the gamma estimator.
Before that, we introduce some notations to simplify the description.
Let
$A^{\otimes(i, j)}$ is an $(i, j)$-element of $A^{\otimes}$,
$A^{\otimes} = A \otimes A = AA^{\star}$, and
$^\star$ denotes transpose.
Let $x\tilde{\otimes}y=((x_iy_j+x_jy_i)/2)\in\mathbb{R}^{\sf r}\otimes\mathbb{R}^{\sf r}$
for $x=(x_i)$, $y=(y_i)\in\mathbb{R}^{\sf r}$, and
let $x^{\tilde{\otimes}(\alpha,\beta)}=x^\alpha_\cdot\tilde{\otimes}x^\beta_\cdot$ for $x=(x^\alpha_i)\in\mathbb{R}^2\otimes\mathbb{R}^{\sf r}$.
We write $x\cdot y=\sum_{i=1}^{\sf r}x_iy_i$ for $x=(x_i)$, $y=(y_i)\in\mathbb{R}^{\sf r}$,
and $x\cdot y = \sum_{i,j=1}^{\sf r}x_{i,j}y_{i,j}$ for
$x=(x_{i,j})$, $y=(y_{i,j})\in\mathbb{R}^{\sf r}\otimes \mathbb{R}^{\sf r}$.
Write
\begin{align*}
\tilde{V}_{k} := \frac{\Delta_{k}V}{a_n h_{k}} = \frac{\Delta_{k}V}{a_n \delta_n}.
\end{align*}
for a stochastic process $V$.
For example, we can write
\begin{align*}
S_n^{\alpha \beta}
= \sum_{j=2}^{b_n} (\tilde{\mathbb{Y}}^n_{j} - \tilde{\mathbb{Y}}^n_{j-1})^{\otimes(\alpha, \beta)},
\quad
\gamma ^{p q} = \int_0^1 (\mathbb{X}_s^{1})^{\tilde{\otimes} p} \cdot
(\mathbb{X}_s^{1})^{\tilde{\otimes}q} \ ds.
\end{align*}
\subsection{$\Gamma$ estimator}
%Following
In the similar way in
Barndorff-Nielsen and Shephard \cite{10.2307/3598838},
%Barndorff - Nielsen ``Econometric analysis of realized covariation: high frequency regression and correlation in financial economics'',
we define
%treats a consistent estimator of multivariate quarticity.
%We borrow the quarticity estimator.
%\begin{eqnarray*}
%H = \sum_{j=1}^{M} x_{j} x_{j}' - \frac{1}{2} \sum_{j=1}^{M-1} (x_{j} x_{j+1}' + x_{j} x_{j+1}'),
%\end{eqnarray*}
%where $x_{j} = vec(y_{j} y_{j}')$.
\begin{align*}
\hat{\gamma}_{n, 1}^{pq}
&=\frac{9}{8} \Big\{ \sum_{k=2}^{b_n}
(\tilde{\mathbb{Y}}^n_{k} - \tilde{\mathbb{Y}}^n_{k-1})^{\otimes p}
(\tilde{\mathbb{Y}}^n_{k} - \tilde{\mathbb{Y}}^n_{k-1})^{\otimes q} \\
&\qquad
- \frac{1}{2} \sum_{k=2}^{b_n - 2}
\Big( (\tilde{\mathbb{Y}}^n_{k} - \tilde{\mathbb{Y}}^n_{k-1})^{\otimes p}
(\tilde{\mathbb{Y}}^n_{k+2} - \tilde{\mathbb{Y}}^n_{k+1})^{\otimes q} \\
&\qquad\quad
+ (\tilde{\mathbb{Y}}^n_{k+2} - \tilde{\mathbb{Y}}^n_{k+1})^{\otimes p}
(\tilde{\mathbb{Y}}^n_{k} - \tilde{\mathbb{Y}}^n_{k-1})^{\otimes q} \Big) \Big\} \Big(\frac{T}{b_n} \Big)^{-1}
\end{align*}
%
By the similar idea, we define
\begin{align*}
\hat{\gamma}_{n, 2}^{pq}
&= \frac{9}{8} \sum_{k=2}^{b_n-2} \frac{1}{2}
\Big\{ (\tilde{\mathbb{Y}}^n_{k+2} - \tilde{\mathbb{Y}}^n_{k+1})^{\otimes p}
- (\tilde{\mathbb{Y}}^n_{k} - \tilde{\mathbb{Y}}^n_{k-1})^{\otimes p} \Big\} \\
&\qquad\qquad
\times \Big\{ (\tilde{\mathbb{Y}}^n_{k+2} - \tilde{\mathbb{Y}}^n_{k+1})^{\otimes q}
- (\tilde{\mathbb{Y}}^n_{k} - \tilde{\mathbb{Y}}^n_{k-1})^{\otimes q} \Big\} \Big(\frac{T}{b_n} \Big)^{-1}.
\end{align*}
%
%I want to prove this convergence under the same condition of the theorem of the asymptotic mixed normality of $C_n^{1 2}$
%Hence we get the confidence interval.
%
% where for a stochastic process $V$,
% \begin{eqnarray*}
% \tilde{V}_{k} := \frac{\Delta_{k}V}{a_n h_{k}} = \frac{\Delta_{k}V}{a_n \delta_n}.
% \end{eqnarray*}
\subsection{Kernel based $\Gamma$ estimator}
%Following
In the similar way in
Hayashi and Yoshida \cite{hayashi2011nonsynchronous}, we define the kernel based estimator.
\begin{align*}
\partial_{h} \{X^{\alpha} , X^{\beta} \}_{k}
:= \sum_{l=(k - n(h) +1) \vee 2}^{k} (\Delta_{l} X^{\alpha} \Delta_{l} X^{\beta}) h^{-1}.
\end{align*}
where $h = h_n$ is a parameter satisfying $h_n \to 0$ and $h_n^{-1} b_n^{-1} \to 0$,
%$n(h) := \min_{n} \{ t_n \geq h\} \sim h b_n$
and $n(h) := \max_{m} \{ t_m \leq h\} $. %\sim h (T/b_n)^{-1}$
%
%
%\begin{eqnarray*}
%\hat{G}_{n}^{pq}
%= \sum_{k=1}^{b_n} \frac{1}{2} \big\{
%\partial_{h} \{X^{\alpha_1} , X^{\alpha_2} \}_{k} \
%\partial_{h} \{X^{\beta_1} , X^{\beta_2} \}_{k}
%+ \partial_{h} \{X^{\alpha_1} , X^{\beta_2} \}_{k} \
%\partial_{h} \{X^{\beta_1} , X^{\alpha_2} \}_{k} \big\} (\frac{T}{b_n})^2
%\end{eqnarray*}
%$p = (\alpha_1, \beta_1), q = (\alpha_2, \beta_2)$
%
\begin{align*}
\hat{\gamma}_{n, h}^{pq}
=& \frac{9}{8}\sum_{k=2}^{b_n} \Big\{
\partial_{h} \{ \tilde{Y}^{n, \alpha_1} , \tilde{Y}^{n, \alpha_2} \}_{k} \
\partial_{h} \{ \tilde{Y}^{n, \beta_1} , \tilde{Y}^{n, \beta_2} \}_{k} \\
&\qquad
+ \partial_{h} \{ \tilde{Y}^{n, \alpha_1} , \tilde{Y}^{n, \beta_2} \}_{k} \
\partial_{h} \{ \tilde{Y}^{n, \beta_1} , \tilde{Y}^{n, \alpha_2} \}_{k} \Big\} \Big(\frac{T}{b_n} \Big).
\end{align*}
%$\hat{\gamma}_{n, h}^{p q}$ : kernel based Gamma estimator.
%$b_n \hat{\gamma}_{n, h}^{p q} \to^p \gamma^{p q} (n \to \infty)$.
\subsection{$\Xi$ estimator}
Here, we define
\begin{align*}
\hat{\Xi}_{n, *}^{1/2}
= \Big(\frac{1}{\sqrt{S_n^{11} S_n^{22}}}, \frac{-S_n^{12}}{2 \sqrt{(S_n^{11})^3 S_n^{22}}}, \frac{-S_n^{12}}{2 \sqrt{S_n^{11} (S_n^{22})^3}} \Big) \hat{\Gamma}_{n, *}^{1/2}
\end{align*}
and
$\hat{\Xi}_{n, *} = (\hat{\Xi}_{n, *}^{1/2})^{\otimes}$, for $* = 1, 2, h$.
\subsection{consistency of the estimators}
\begin{theorem*} \label{nonkernel}
(a)
$%(T/b_n)^{-1}
\hat{\gamma}_{n, i}^{p q} = \gamma^{p q} + o_p(1)$
under [A] and any one of [B], [B$'$] and [B$^{\sharp^{\sharp}}$], ($i=1, 2$).
(b)
$%(T/b_n)^{-1}
\hat{\gamma}_{n, i}^{p q} = \gamma^{p q} + O_p(b_n^{-1/2})$
under [A] and [B$^{\sharp^{\sharp}}$], ($i=1, 2$).
(c)
$%(T/b_n)^{-1}
\hat{\gamma}_{n, i}^{p q} = \gamma^{p q} + O_p(b_n^{-1/2})$
under [A], [B] or [B'], and [C] (i=1,2).
\end{theorem*}
%
%\begin{corollary*}
%$(T/b_n)^{-1} \hat{\gamma}_{n, i}^{p q} \to^p \gamma^{p q}$
%under [A] and any one of [B], [B$'$] and [B$^{\sharp^{\sharp}}$], ($i=1, 2$).
%\end{corollary*}
%
\begin{theorem*} \label{kernel}
(a)
$%(T/b_n)^{-1}
\hat{\gamma}_{n, h}^{p q} = \gamma^{p q} + o_p(1)$
under [A] and any one of [B], [B$'$] and [B$^{\sharp^{\sharp}}$]. %, ($i=1, 2$).
(b)
$%(T/b_n)^{-1}
\hat{\gamma}_{n, h}^{p q} = \gamma^{p q} + O_p(b_n^{-1/2}) + O_p(h_n)$
under %$h_n = O(b_n^{-1/2})$,
[A] and [B$^{\sharp^{\sharp}}$]. %, ($i=1, 2$).
(c)
$%(T/b_n)^{-1}
\hat{\gamma}_{n, h}^{p q} = \gamma^{p q} + O_p(b_n^{-1/2}) + O_p(h_n)$
under %$h_n = O(b_n^{-1/2})$,
[A], [B] or [B'], and [C]. % (i=1,2).
%$%(T/b_n)^{-1}
%\hat{\gamma}_{n, h}^{p q} \to^p \gamma^{p q}$
%under [A] and any one of [B], [B$'$] and [B$^{\sharp^{\sharp}}$].
\end{theorem*}
\begin{corollary*} \label{avar}
(a)
$%(T/b_n)^{-1}
\hat{\Xi}_{n, i} = \Xi + o_p(1)$
under [A] and any one of [B], [B$'$] and [B$^{\sharp^{\sharp}}$], ($i=1, 2$).
(b)
$%(T/b_n)^{-1}
\hat{\Xi}_{n, i} = \Xi + O_p(b_n^{-1/2})$
under [A] and [B$^{\sharp^{\sharp}}$], ($i=1, 2$).
(c)
$%(T/b_n)^{-1}
\hat{\Xi}_{n, i} = \Xi + O_p(b_n^{-1/2})$
under [A], [B] or [B'], and [C] (i=1,2).
\end{corollary*}
\begin{corollary*} \label{avar_k}
(a)
$%(T/b_n)^{-1}
\hat{\Xi}_{n, h} = \Xi + o_p(1)$
under [A] and any one of [B], [B$'$] and [B$^{\sharp^{\sharp}}$]. %, ($i=1, 2$).
(b)
$%(T/b_n)^{-1}
\hat{\Xi}_{n, h} = \Xi + O_p(b_n^{-1/2}) + O_p(h_n)$
under %$h_n = O(b_n^{-1/2})$,
[A] and [B$^{\sharp^{\sharp}}$]. %, ($i=1, 2$).
(c)
$%(T/b_n)^{-1}
\hat{\Xi}_{n, h} = \Xi + O_p(b_n^{-1/2}) + O_p(h_n)$
under %$h_n = O(b_n^{-1/2})$,
[A], [B] or [B'], and [C]. % (i=1,2).
\end{corollary*}
\section{Proof} \label{proof}
\subsection*{Proof of Theorem \ref{nonkernel}}
%
First, we approximate $\hat{\gamma}_{n, i}^{p q} (i = 1, 2)$ by the following quantities.
Let
\begin{align*}
G_{n,1}^{pq}
&= \frac{9}{8} \Big\{ \sum_{k=2}^{b_n}
(\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1})^{\otimes p}
(\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1})^{\otimes q} \\
&\qquad
- \frac{1}{2} \sum_{k=2}^{b_n - 2}
\Big( (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1})^{\otimes p}
(\tilde{\bm{\chi}}^n_{k+2} - \tilde{\bm{\chi}}^n_{k+1})^{\otimes q} \\
&\qquad\qquad\qquad
+ (\tilde{\bm{\chi}}^n_{k+2} - \tilde{\bm{\chi}}^n_{k+1})^{\otimes p}
(\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1})^{\otimes q} \Big) \Big\} \Big(\frac{T}{b_n} \Big)^{-1}
\end{align*}
and
\begin{align*}
G_{n,2}^{pq}
&= \frac{9}{8} \sum_{k=2}^{b_n-2} \frac{1}{2}
\Big\{ (\tilde{\bm{\chi}}^n_{k+2} - \tilde{\bm{\chi}}^n_{k+1})^{\otimes p}
- (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1})^{\otimes p} \Big\} \\
&\qquad\qquad
\times
\Big\{ (\tilde{\bm{\chi}}^n_{k+2} - \tilde{\bm{\chi}}^n_{k+1})^{\otimes q}
- (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1})^{\otimes q} \Big\} \Big(\frac{T}{b_n} \Big)^{-1}. %,
\end{align*}
where
$\bm{\chi}^n = (\chi^{n, 1}, \chi^{n, 2})$ and
$\chi^{n, \alpha}_t = \int_{0}^{t} a_n X^{\alpha}_{s} \ ds$.
%
% 20170807
% \mathbb{\chi} is replaced by \bm{\chi} because \mathbb{\chi} equals \chi.
%
\begin{lemma*} \label{lemma1i}
(a)
$%(T/b_n)^{-1}
\hat{\gamma}^{p q}_{n,i} = %(T/b_n)^{-1}
G_{n,i}^{p q} + o_p(1)$
under any one of [B], [B$'$] and [B$^{\sharp^{\sharp}}$], (i=1,2).
(b)
$%(T/b_n)^{-1}
\hat{\gamma}^{p q}_{n,i} = %(T/b_n)^{-1}
G_{n,i}^{p q} + o_p(b_n^{-1/2})$
under [B$^{\sharp^{\sharp}}$], (i=1,2).
\end{lemma*}
\proof
Write
$(V^1 \otimes V^2 \otimes \cdots \otimes V^n)_{i_1, i_2, \ldots, i_n} = V^1_{i_1} V^2_{i_2} \cdots V^n_{i_n}$
for vectors $V^1, V^2, \ldots, V^n$
and confirm
$|V^1 \otimes V^2 \otimes \cdots \otimes V^n | = |V^1| |V^2| \cdots |V^n|$.
Let
$s_{k} = (\tilde{\mathbb{Y}}^n_{k} - \tilde{\mathbb{Y}}^n_{k-1})^{\otimes}$,
$t_{k} = (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1})^{\otimes}$,
%
%$s_{k}^{\alpha \beta} = (\tilde{\mathbb{Y}}^n_{k} - \tilde{\mathbb{Y}}^n_{k-1})^{\otimes (\alpha,\beta)}$,
%$t_{k}^{\alpha \beta} = (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1})^{\otimes (\alpha,\beta)}$,
%$s_{k} = (s_{k}^{12}, s_{k}^{21}, s_{k}^{11}, s_{k}^{22})^{\star}$,
%$t_{k} = (t_{k}^{12}, t_{k}^{21}, t_{k}^{11}, t_{k}^{22})^{\star}$,
%
%\begin{eqnarray*}
%\check{s}_{k, h}%^{\alpha \beta}
%&:=& \sum_{l=(k - n(h) +1)_{+}}^{k} s_{l}%^{\alpha \beta}
%\ h^{-1} \\
%\check{t}_{k, h}%^{\alpha \beta}
%&:=& \sum_{l=(k - n(h) +1)_{+}}^{k} t_{l}%^{\alpha \beta}
%\ h^{-1}
%\end{eqnarray*}
%$\check{s}_{k, h} = (\check{s}_{k, h}^{12}, \check{s}_{k, h}^{11}, \check{s}_{k, h}^{22})^{\star}$,
%$\check{t}_{k, h} = (\check{t}_{k, h}^{12}, \check{t}_{k, h}^{11}, \check{t}_{k, h}^{22})^{\star}$. 0
%
$\hat{\gamma}_{n,i}
= (\hat{\gamma}^{\alpha_1, \beta_1, \alpha_2, \beta_2}_{n,i})_{\alpha_1, \beta_1, \alpha_2, \beta_2 = 1, 2}$,
and
${G}_{n,i}
= ({G}^{\alpha_1, \beta_1, \alpha_2, \beta_2}_{n,i})_{\alpha_1, \beta_1, \alpha_2, \beta_2 = 1, 2}$ ($i=1, 2$).
%
%$\hat{\gamma}_{n,i} = (\hat{\gamma}^{p q}_{n,i})_{p, q = (1,2),(2,1),(1,1),(2,2)}$, and
%${G}_{n,i} = ({G}^{p q}_{n,i})_{p, q = (1,2),(2,1),(1,1),(2,2)}$ ($i=1, 2$).
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} | \hat{\gamma}_n - G_n |
%&=& 2 (\frac{T}{b_n})^{-1} | \sum_{k=2}^{b_n} (\check{s}_{k, h} \cdot \check{s}_{k, h}^{\star} - \check{t}_{k, h} \cdot \check{t}_{k, h}^{\star}) | (\frac{T}{b_n})^{2} \\
%&=& 2 (\frac{T}{b_n})^{-1} | \sum_{k=2}^{b_n} (\check{s}_{k, h} - \check{t}_{k, h}) \cdot (\check{s}_{k, h} - \check{t}_{k, h})^{\star} \\
%&&+ (\check{s}_{k, h} - \check{t}_{k, h}) \cdot \check{t}_{k, h}^{\star}
%+ \check{t}_{k, h} \cdot (\check{s}_{k, h} - \check{t}_{k, h})^{\star} | (\frac{T}{b_n})^{2} \\
%&\leq& 2 (\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n} \{ | (\check{s}_{k, h} - \check{t}_{k, h}) \cdot (\check{s}_{k, h} - \check{t}_{k, h})^{\star} | \\
%&&+ | (\check{s}_{k, h} - \check{t}_{k, h}) \cdot \check{t}_{k, h}^{\star} |
%+ | \check{t}_{k, h} \cdot (\check{s}_{k, h} - \check{t}_{k, h})^{\star} | \} (\frac{T}{b_n})^{2} \\
%&=& 2 (\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n} | \check{s}_{k, h} - \check{t}_{k, h} |^2 (\frac{T}{b_n})^{2}
%+ 4 (\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n} | \check{t}_{k, h} | | \check{s}_{k, h} - \check{t}_{k, h} | (\frac{T}{b_n})^{2} \\
%&\leq& 2 ((\frac{T}{b_n})^{-1/2} \sum_{k=2}^{b_n} | \check{s}_{k, h} - \check{t}_{k, h} | (\frac{T}{b_n}))^2 \\
%&&+ 4 (\frac{T}{b_n})^{-1/2} \max_{k} | (\frac{T}{b_n}) \check{t}_{k, h} | ((\frac{T}{b_n})^{-1/2} \sum_{k=2}^{b_n} | \check{s}_{k, h} - \check{t}_{k, h} | (\frac{T}{b_n}))
%\end{eqnarray*}
It holds that
\begin{align*}
&\Big(\frac{9}{8} \Big)^{-1} %\Big(\frac{T}{b_n} \Big)^{-1}
| \hat{\gamma}_{n,1} - G_{n,1} | \\
&= \Big(\frac{T}{b_n} \Big)^{-1} \Big| \sum_{k=2}^{b_n} (s_{k} \otimes s_{k} - t_{k} \otimes t_{k}) \\
&\qquad\qquad\quad
- \frac{1}{2} \sum_{k=2}^{b_n - 2} \Big( (s_{k} \otimes s_{k+2} - t_{k} \otimes t_{k+2})
+ (s_{k+2} \otimes s_{k} - t_{k+2} \otimes t_{k}) \Big) \Big| \\
&\leq \Big(\frac{T}{b_n} \Big)^{-1} \Big\{ \sum_{k=2}^{b_n} |s_{k} \otimes s_{k} - t_{k} \otimes t_{k}| \\
&\qquad\qquad\quad
+ \frac{1}{2} \sum_{k=2}^{b_n - 2} |s_{k} \otimes s_{k+2} - t_{k} \otimes t_{k+2}|
+ \frac{1}{2} \sum_{k=2}^{b_n - 2} |s_{k+2} \otimes s_{k} - t_{k+2} \otimes t_{k}| \Big\}
\end{align*}
and
\begin{align*}
&\Big(\frac{9}{8} \Big)^{-1} %\Big(\frac{T}{b_n}\Big)^{-1}
| \hat{\gamma}_{n,2} - G_{n,2} | \\
&= \frac{1}{2} \Big(\frac{T}{b_n} \Big)^{-1} \Big| \sum_{k=2}^{b_n-2}
(s_{k+2} \otimes s_{k+2} - t_{k+2} \otimes t_{k+2})
- (s_{k+2} \otimes s_{k} - t_{k+2} \otimes t_{k}) \\
&\qquad\qquad\qquad\quad
- (s_{k} \otimes s_{k+2} - t_{k} \otimes t_{k+2})
+ (s_{k} \otimes s_{k} - t_{k} \otimes t_{k}) \Big| \\
&\leq \frac{1}{2} \Big(\frac{T}{b_n} \Big)^{-1} \Big\{ \sum_{k=2}^{b_n-2}
|s_{k+2} \otimes s_{k+2} - t_{k+2} \otimes t_{k+2}|
+ \sum_{k=2}^{b_n-2} |s_{k+2} \otimes s_{k} - t_{k+2} \otimes t_{k}| \\
&\qquad\qquad\qquad
+ \sum_{k=2}^{b_n-2} |s_{k} \otimes s_{k+2} - t_{k} \otimes t_{k+2}|
+ \sum_{k=2}^{b_n-2} |s_{k} \otimes s_{k} - t_{k} \otimes t_{k}| \Big\}.
\end{align*}
Now, let $\rho=0, 1/2$ and see the typical term.
\begin{align*}
&b_n^{1+\rho} \sum_{k=2}^{b_n-2}
|s_{k+2} \otimes s_{k} - t_{k+2} \otimes t_{k}| \\
&= b_n^{1+\rho} \sum_{k=2}^{b_n-2}
\Big|(s_{k+2} - t_{k+2}) \otimes (s_{k} - t_{k}) \\
&\qquad\qquad\quad
+ (s_{k+2} - t_{k+2}) \otimes t_{k}
+ t_{k+2} \otimes (s_{k} - t_{k}) \Big| \\
&\leq b_n^{1+\rho} \sum_{k=2}^{b_n-2}
\Big|(s_{k+2} - t_{k+2}) \otimes (s_{k} - t_{k}) \Big| \\
&\qquad\qquad\quad
+ \Big|(s_{k+2} - t_{k+2}) \otimes t_{k} \Big|
+ \Big|t_{k+2} \otimes (s_{k} - t_{k}) \Big| \\
&= b_n^{1+\rho} \sum_{k=2}^{b_n-2}
|s_{k+2} - t_{k+2}| |s_{k} - t_{k}|
+ |s_{k+2} - t_{k+2}| |t_{k}|
+ |t_{k+2}| |s_{k} - t_{k}| \\
%&&b_n^{3/2} \sum_{k=2}^{b_n-2}
%|s_{k+2} \otimes s_{k} - t_{k+2} \otimes t_{k}| \\
&\leq \Big(b_n^{1/2 + \rho/2} \sum_{k=2}^{b_n-2} |s_{k+2} - t_{k+2}| \Big)
\Big(b_n^{1/2 + \rho/2} \sum_{k=2}^{b_n-2} |s_{k} - t_{k}| \Big) \\
&\quad
+ \Big(b_n^{\rho} \sum_{k=2}^{b_n-2} |s_{k+2} - t_{k+2}|\Big) \Big(b_n \max_{k} |t_{k}| \Big) \\
&\quad
+ \Big(b_n \max_{k} |t_{k+2}| \Big) \Big(b_n^{\rho} \sum_{k=2}^{b_n-2} |s_{k} - t_{k}| \Big).
%
%
%&\leq& (\sqrt{b_n} \sum_{k=2}^{b_n-2} |s_{k+2} - t_{k+2}|) (\sqrt{b_n} \sum_{k=2}^{b_n-2} |s_{k} - t_{k}|) \\
%&&+ (b_n \sum_{k=2}^{b_n-2} |s_{k+2} - t_{k+2}|^2)^{1/2} (b_n \sum_{k=2}^{b_n-2} |t_{k}|^2)^{1/2}
%+ (b_n \sum_{k=2}^{b_n-2} |t_{k+2}|^2)^{1/2} (b_n \sum_{k=2}^{b_n-2} |s_{k} - t_{k}|^2)^{1/2}
\end{align*}
Under [B$^{\sharp^{\sharp}}$], %we have
the same argument of the proof of Lemma 1 (b) in \cite{Kimura2016} under the assumption [B] yields
\begin{align*}
b_n^{3/4} \sum_{k=2}^{b_n} |s_{k} - t_{k}| \to^p 0.
\end{align*}
%by the same argument of the proof of Lemma 1 (b) in \cite{Kimura2016} under the assumption [B].
Under any one of [B], [B$'$] and [B$^{\sharp^{\sharp}}$], it holds that
\begin{align*}
b_n^{1/2} \sum_{k=2}^{b_n} |s_{k} - t_{k}| \to^p 0
\end{align*}
by the proof of Lemma 1 (b) in \cite{Kimura2016}.
By inequality (4) in \cite{Kimura2016}, and chebychev's inequality or Markov's inequality, we have
\begin{align*}
b_n \max_{k} |t_{k}| = O_p(1).
\end{align*}
Therefore, the typical term converges to $0$ in probability.
The other terms are evaluated in the same way.
We obtain the conclusion.
%
%
%By proof of Lemma 1. and the inequalities
%\begin{eqnarray*}
%\sum_{k=2}^{b_n} | \check{s}_{k, h} - \check{t}_{k, h} | (\frac{T}{b_n})
%\leq \sum_{k=2}^{b_n} \sum_{l=(k - n(h) +1)_{+}}^{k} | s_{l} - t_{l} | \ h^{-1} (\frac{T}{b_n})
%\leq \sum_{k=2}^{b_n} | s_{k} - t_{k} |,
%\end{eqnarray*}
%\begin{eqnarray*}
%\max_{k} | (\frac{T}{b_n}) \check{t}_{k, h} |
%\leq \max_{k} | t_{k} |,
%\end{eqnarray*}
%(4) and
%\begin{eqnarray*}
%|S_n - T_n|
%\leq \sum_{k=2}^{b_n} | s_{k} - t_{k} |
%\leq \sum_{k=2}^{b_n} | \tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1}|^2
%+ 2 \{ \sum_{k=2}^{b_n} | \tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1}|^2 \}^{1/2}
%\{ \sum_{k=2}^{b_n} | \tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1}|^2 \}^{1/2},
%\end{eqnarray*}
%we get the conclusion.
\qed
%
%Under [B$^{\sharp^{\sharp}}$], we can prove
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} \hat{\gamma}^{p q}_n = (\frac{T}{b_n})^{-1} G^{p q}_{n, i} + o_p(b_n^{-1/2}).
%\end{eqnarray*}
%\begin{eqnarray*}
%&&b_n^{3/2} \sum_{k=2}^{b_n-2}
%|s_{k+2} \cdot s_{k}^{\star} - t_{k+2} \cdot t_{k}^{\star}| \\
%&\leq& (b_n^{3/4} \sum_{k=2}^{b_n-2} |s_{k+2} - t_{k+2}|) (b_n^{3/4} \sum_{k=2}^{b_n-2} |s_{k} - t_{k}|) \\
%&&+ (b_n^{1/2} \sum_{k=2}^{b_n-2} |s_{k+2} - t_{k+2}|) (b_n \max_{k} |t_{k}|)
%+ (b_n \max_{k} |t_{k+2}|) (b_n^{1/2} \sum_{k=2}^{b_n-2} |s_{k} - t_{k}|)
%\end{eqnarray*}
%By similar (same) argument of the proof of Lemma 1. (b) under the assumption [B], we have
%\begin{eqnarray*}
%b_n^{3/4} \sum_{k=2}^{b_n} |s_{k} - t_{k}| \to^p 0.
%\end{eqnarray*}
%By inequality (4), and chebychev's inequality or Markov's inequality, we have
%\begin{eqnarray*}
%b_n \max_{k} |t_{k}| = O_p(1).
%\end{eqnarray*}
%
%\begin{lemma*}
%$\sum_{l=(k - n(h) +1)_{+}}^{k} F_{l} G_{l} h^{-1} \to^p 0$
%for all pairs of
%$(F_{l}^, G_{l}) \in \{ A_{k}^{\alpha}, B_{k}^{\alpha}, C_{k}^{\alpha}, D_{k}^{\alpha} ; \alpha=1, 2 \}^2 \backslash \{(A_{l}^{\alpha}, A_{l}^{\beta}) ; \alpha, \beta =1, 2 \}$.
%\end{lemma*}
%
%The proof of this lemma is very similar to proof of Lemma 2.
%
%\proof
%Let
%$\mathcal{Q}_{n, h} = \sum_{l=(k - n(h) +1)_{+}}^{k} F_{l} G_{l} h^{-1}$.
%
%For $F_{l} = A_{l}^{\alpha}$ and $G_{l} = C_{l}^{\beta}$,
%$E[|\mathcal{Q}_{n, h}|] = O(n(h) h^{-1} b_n^{-3/2}) = O(b_n^{-1/2})$.
%
%For $F_{l} = A_{l}^{\alpha}$ and $G_{l} = D_{l}^{\beta}$,
%$E[|\mathcal{Q}_{n, h}|] = O(n(h) h^{-1} b_n^{-3/2}) o(1) = o(b_n^{-1/2})$.
%
%%
%%For $F_{l} = A_{l}^{\alpha}$ and $B_{l} = C_{l}^{\beta}$,
%%$E[|\mathcal{Q}_{n, h}|^2] = O(n(h) h^{-2} b_n^{-3}) = O(b_n^{-2} h^{-1})$.
%%Therefore, $\mathcal{Q}_{n, h} = O_p(b_n^{-1/2} (h^{-1} b_n^{-1/2})) = O_p(b_n^{-1/2})$.
%%
%%
%For $F_{l} = A_{l}^{\alpha}$ and $G_{l} = B_{l}^{\beta}$,
%$E[|\mathcal{Q}_{n, h}|] = O(n(h) h^{-1} b_n^{-1}) o(1) = o(1)$.
%
%\qed
%
%
%\begin{eqnarray*}
%Q_{k, h}^{1, \alpha \beta} = \sum_{l=(k - n(h) +1)_{+}}^{k} A_{l}^{\alpha} A_{l}^{\beta} h^{-1}
%\end{eqnarray*}
%
%By Lemma 5.2. and (4)
%(esp. $Q_{k, h}^{1, \alpha \beta} = O_p(1)$),
%we have
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} G_{n}^{p q}
%= (\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n} \frac{1}{2} (Q_{k, h}^{1, \alpha_1 \alpha_2} Q_{k, h}^{1, \beta_1 \beta_2}
%+ Q_{k, h}^{1, \alpha_1 \beta_2} Q_{k, h}^{1, \beta_1 \alpha_2}) (\frac{T}{b_n})^{2} + o_p(1)
%=: b_n \mathcal{R}_{n}^{1, p q} + o_p(1).
%\end{eqnarray*}
%
\begin{proposition*} \label{prop1i}
$%(T/b_n)^{-1}
\hat{\gamma}^{p q}_{n,i} = %(T/b_n)^{-1}
G_{n,i}^{p q} + o_p(b_n^{-1/2})$
under [B] or [B'], and [C] (i=1,2).
\end{proposition*}
%
\proof
We just confirm the convergence of the easiest term.
\begin{align*}
& b_n^{1+\rho} \sum_{k=2}^{b_n-2} |s_{k} \otimes s_{k} - t_{k} \otimes t_{k}| \\
&= b_n^{1+\rho} \sum_{k=2}^{b_n-2}
\Big|(s_{k} - t_{k}) \otimes (s_{k} - t_{k})
+ (s_{k} - t_{k}) \otimes t_{k}
+ t_{k} \otimes (s_{k} - t_{k}) \Big|
\end{align*}
By definition,
\begin{align*}
s_k - t_k
&= (\tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1})^{\otimes}
+ (\tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1})
\otimes (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1}) \\
&\quad
+ (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1})
\otimes (\tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1}).
\end{align*}
Therefore, it is enough to prove the following equations.
\begin{align*}
b_n^{1+\rho} \sum_{k=2}^{b_n-2}
(\tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1})^{\otimes 3}
&=o_p(1)\\
b_n^{1+\rho} \sum_{k=2}^{b_n-2}
(\tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1})^{\otimes 2}
\otimes (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1})
&=o_p(1)\\
b_n^{1+\rho} \sum_{k=2}^{b_n-2}
(\tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1})^{\otimes}
\otimes (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1})^{\otimes}
&=o_p(1)\\
b_n^{1+\rho} \sum_{k=2}^{b_n-2}
(\tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1})
\otimes (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1})^{\otimes 2}
&=o_p(1),
\end{align*}
where for vector $V$,
$V^{\otimes} = V \otimes V$,
$V^{\otimes 2} = V \otimes V \otimes V$,
$V^{\otimes 3} = V \otimes V \otimes V \otimes V$.
By assumption [C], it holds that
\begin{align*}
E\bigg[ \bigg| b_n^{1+\rho} \sum_{k=2}^{b_n-2}
(\tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1})^{\otimes 3} \bigg| \bigg]
&\leq
E\bigg[ b_n^{1+\rho} \sum_{k=2}^{b_n-2}
| \tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1} |^{4} \bigg] \\
&\leq
16 E\bigg[ b_n^{1+\rho} \sum_{k=2}^{b_n-2}
| \tilde{\mathbb{M}}^n_{k} |^{4} \bigg] \\
&= O(b_n^{4+\rho} a_n^{-2})
= o(1).
\end{align*}
%By Lemma 1 (b) and inequality (4) in \cite{Kimura2016}, we have
Lemma 1 (b) and inequality (4) in \cite{Kimura2016} yields
\begin{align*}
&\bigg| b_n^{1+\rho} \sum_{k=2}^{b_n-2}
(\tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1})^{\otimes}
\otimes (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1})^{\otimes} \bigg| \\
&\leq
\bigg( b_n^{\rho} \sum_{k=2}^{b_n-2}
| \tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1} |^2 \bigg)
\bigg( b_n |\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1} |^2 \bigg)
= o_p(1) O_p(1)
= o_p(1).
\end{align*}
and
\begin{align*}
&\bigg| b_n^{1+\rho} \sum_{k=2}^{b_n-2}
(\tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1})
\otimes (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1})^{\otimes \otimes} \bigg| \\
&\leq
\bigg| b_n^{\rho} \sum_{k=2}^{b_n-2}
(\tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1}) \otimes (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1}) \bigg|
\bigg| b_n (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1} )^{\otimes} \bigg| \\
&= o_p(1) O_p(1)
= o_p(1).
\end{align*}
From the above inequalities, it is derived that
\begin{align*}
&\bigg| b_n^{1+\rho} \sum_{k=2}^{b_n-2}
(\tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1})^{\otimes 2}
\otimes (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1}) \bigg| \\
&\leq
\bigg| %\bigg\{
b_n^{1+\rho} \sum_{k=2}^{b_n-2}
(\tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1})^{\otimes 3} %\bigg\}
\bigg|^{1/2} \\
&\qquad
\times
%\bigg\{
\bigg| b_n^{1+\rho} \sum_{k=2}^{b_n-2}
(\tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1})^{\otimes}
\otimes (\tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1})^{\otimes} %\bigg\}
\bigg|^{1/2} \\
&= o_p(1) o_p(1) = o_p(1).
\end{align*}
\qed
Now, we let
\begin{align*}
A^\alpha_k &=
X_{t_{k-2}}^{\alpha 1} \int_{I_{k-1}} H_k(s) \ dw_s +X_{t_{k-2}}^{\alpha 1} \int_{I_{k}} K_k(s) \ dw_s \\
B^\alpha_k &= \frac{1}{h_{k}} \int_{I_{k}} \int_{t_{k-2}}^{t} (X_{s}^{\alpha 1}-X_{t_{k-2}}^{\alpha 1}) \ dw_s \ dt - \frac{1}{h_{k-1}} \int_{I_{k-1}}\int_{t_{k-2}}^{t}(X_{s}^{\alpha 1}-X_{t_{k-2}}^{\alpha 1}) \ dw_s \ dt
\\
C^\alpha_k &= \frac{1}{h_{k}} \int_{I_{k}} \int_{t_{k-2}}^{t} X_{t_{k-2}}^{\alpha 0}ds \ dt - \frac{1}{h_{k-1}} \int_{I_{k-1}}\int_{t_{k-2}}^{t}X_{t_{k-2}}^{\alpha 0}ds \ dt
\\
D^\alpha_k &= \frac{1}{h_{k}} \int_{I_{k}} \int_{t_{k-2}}^{t} (X_{s}^{\alpha 0}-X_{t_{k-2}}^{\alpha 0})ds \ dt - \frac{1}{h_{k-1}} \int_{I_{k-1}}\int_{t_{k-2}}^{t}(X_{s}^{\alpha 0}-X_{t_{k-2}}^{\alpha 0})ds \ dt.
\end{align*}
By Lemma 2 and inequality (4) in \cite{Kimura2016}, we have
\begin{align*}
%\Big(\frac{T}{b_n} \Big)^{-1}
G_{n,1}^{p q}
&= \frac{9}{8} \Big\{ \sum_{k=2}^{b_n}
(A^{\alpha_1}_{k} A^{\beta_1}_{k}) (A^{\alpha_2}_{k} A^{\beta_2}_{k}) \\
&\qquad\qquad\qquad
- \frac{1}{2} \sum_{k=2}^{b_n-2}
\Big( (A^{\alpha_1}_{k} A^{\beta_1}_{k}) (A^{\alpha_2}_{k+2} A^{\beta_2}_{k+2}) \\
&\qquad\qquad\qquad\qquad\qquad
+ (A^{\alpha_1}_{k+2} A^{\beta_1}_{k+2}) (A^{\alpha_2}_{k} A^{\beta_2}_{k}) \Big)
\Big\} \Big(\frac{T}{b_n} \Big)^{-1}
+ o_p(b_n^{-1/2}) \\
&= %\Big(\frac{T}{b_n} \Big)^{-1}
\mathcal{R}_{n,1}^{1,p q} + o_p(b_n^{-1/2})
\end{align*}
and
\begin{align*}
%\Big(\frac{T}{b_n} \Big)^{-1}
G_{n,2}^{p q}
&= \frac{9}{8} \sum_{k=2}^{b_n-2} \frac{1}{2} \Big\{
(A^{\alpha_1}_{k+2} A^{\beta_1}_{k+2}) (A^{\alpha_2}_{k+2} A^{\beta_2}_{k+2}) \\
&\qquad\qquad\qquad\qquad\quad
- (A^{\alpha_1}_{k+2} A^{\beta_1}_{k+2}) (A^{\alpha_2}_{k} A^{\beta_2}_{k}) \\
&\qquad\qquad\qquad\qquad\quad
- (A^{\alpha_1}_{k} A^{\beta_1}_{k}) (A^{\alpha_2}_{k+2} A^{\beta_2}_{k+2})\\
&\qquad\qquad\qquad\qquad\quad
+ (A^{\alpha_1}_{k} A^{\beta_1}_{k}) (A^{\alpha_2}_{k} A^{\beta_2}_{k})
\Big\} \Big(\frac{T}{b_n} \Big)^{-1} + o_p(b_n^{-1/2}) \\
&= %\Big(\frac{T}{b_n} \Big)^{-1}
\mathcal{R}_{n,2}^{1,p q} + o_p(b_n^{-1/2}).
\end{align*}
%
Here, we write
\begin{align*}
q^{2,\alpha \beta}_k
&=
% \sum_{j = 2}^{b_n} \bigg\{
% (\bbX_{{\colorr \taut}_{j-2}}^{1})^{{\colorr\tilde{\otimes}}(\alpha,\beta)}
X^{\alpha1}_{t_{k-2}}\cdot X^{\beta1}_{t_{k-2}}
\frac{h_{k-1}}{3}
+
%(\bbX_{{\colorr \taut}_{j-2}}^{1})^{{\colorr\tilde{\otimes}}(\alpha,\beta)}
X^{\alpha1}_{t_{k-2}}\cdot X^{\beta1}_{t_{k-2}}
\frac{h_{k}}{3}%\bigg\}
\end{align*}
and
\begin{align*}
q_k^{3, \alpha \beta}
&=
% \sum_{j = 2}^{b_n}
2 (\mathbb{X}_{t_{k-2}}^{1})^{\tilde{\otimes}(\alpha,\beta)}
\cdot
\Big( \int_{I_{k-1}} \int_{t_{k-2}}^{t} H_k(s) \ dw_s
\otimes H_k(t) \ dw_t \\
&\qquad\qquad\qquad\qquad
+ \int_{I_{k}} \int_{t_{k-1}}^{t} K_k(s) \ dw_s \otimes K_k(t) \ dw_t \\
&\qquad\qquad\qquad\qquad
+ \int_{I_{k-1}} H_k(s) \ dw_s \otimes \int_{I_{k}} K_k(s) \ dw_s \Big).
\end{align*}
%
Then, we have
\begin{align*}
%\Big(\frac{T}{b_n} \Big)^{-1}
\mathcal{R}_{n,1}^{1,p q}
&= \frac{9}{8} \Big\{ \sum_{k=2}^{b_n}
(q_{k}^{2, p} + q_{k}^{3, p}) (q_{k}^{2, q} + q_{k}^{3, q}) \\
&\qquad\qquad\qquad
- \frac{1}{2} \sum_{k=2}^{b_n-2}
\Big( (q_{k}^{2, p} + q_{k}^{3, p}) (q_{k+2}^{2, q} + q_{k+2}^{3, q}) \\
&\qquad\qquad\qquad\qquad\qquad
+ (q_{k+2}^{2, p} + q_{k+2}^{3, p}) (q_{k}^{2, q} + q_{k}^{3, q}) \Big)
\Big\} \Big(\frac{T}{b_n} \Big)^{-1}
\end{align*}
and
\begin{align*}
%\Big(\frac{T}{b_n} \Big)^{-1}
\mathcal{R}_{n,2}^{1,p q}
&= \frac{9}{8} \sum_{k=2}^{b_n-2} \frac{1}{2} \Big\{
(q_{k+2}^{2, p} + q_{k+2}^{3, p}) (q_{k+2}^{2, p} + q_{k+2}^{3, p}) \\
&\qquad\qquad\qquad\qquad
- (q_{k+2}^{2, p} + q_{k+2}^{3, p}) (q_{k}^{2, p} + q_{k}^{3, p}) \\
&\qquad\qquad\qquad\qquad
- (q_{k}^{2, p} + q_{k}^{3, p}) (q_{k+2}^{2, p} + q_{k+2}^{3, p}) \\
&\qquad\qquad\qquad\qquad
+ (q_{k}^{2, p} + q_{k}^{3, p}) (q_{k}^{2, p} + q_{k}^{3, p})
\Big\} \Big(\frac{T}{b_n} \Big)^{-1}
\end{align*}
by It\^o's formula and simple calculus.
%
We only see the typical terms.
%It holds that
We have
\begin{align} \label{2k2k}
\Big(\frac{T}{b_n} \Big)^{-1} \sum_{k=2}^{b_n} q_{k}^{2, p} q_{k}^{2, q}
= \frac{4}{9} \sum_{k=2}^{b_n} (X_{t_{k-2}}^{\alpha_1 1} \cdot X_{t_{k-2}}^{\beta_1 1})
(X_{t_{k-2}}^{\alpha_2 1} \cdot X_{t_{k-2}}^{\beta_2 1}) \Big(\frac{T}{b_n} \Big),
\end{align}
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n-2} q_{k+2}^{2, p} q_{k}^{2, q}
%= \frac{4}{9} \sum_{k=2}^{b_n-2} (X_{t_{k}}^{\alpha_1 1} \cdot X_{t_{k}}^{\beta_1 1})
%(X_{t_{k-2}}^{\alpha_2 1} \cdot X_{t_{k-2}}^{\beta_2 1}) \frac{T}{b_n}
%\end{eqnarray*}
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n-2} q_{k}^{2, p} q_{k+2}^{2, q}
%= \frac{4}{9} \sum_{k=2}^{b_n-2} (X_{t_{k-2}}^{\alpha_1 1} \cdot X_{t_{k-2}}^{\beta_1 1})
%(X_{t_{k}}^{\alpha_2 1} \cdot X_{t_{k}}^{\beta_2 1}) \frac{T}{b_n}
%\end{eqnarray*}
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n-2} q_{k+2}^{2, p} q_{k+2}^{2, q}
%= \frac{4}{9} \sum_{k=2}^{b_n-2} (X_{t_{k}}^{\alpha_1 1} \cdot X_{t_{k}}^{\beta_1 1})
%(X_{t_{k}}^{\alpha_2 1} \cdot X_{t_{k}}^{\beta_2 1}) \frac{T}{b_n}
%\end{eqnarray*}
%
\begin{align} \label{2k3k}
\Big(\frac{T}{b_n} \Big)^{-1} \sum_{k=2}^{b_n} q_{k}^{2, p} q_{k}^{3, q}
=\sum_{k=2}^{b_n} \frac{2}{3} (X_{t_{k-2}}^{\alpha_1 1} \cdot X_{t_{k-2}}^{\beta_1 1}) q_{k}^{3, q}
= O_p(b_n^{-1/2}),
\end{align}
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n-2} q_{k+2}^{2, p} q_{k}^{3, q}
%=\sum_{k=2}^{b_n-2} \frac{2}{3} (X_{t_{k}}^{\alpha_1 1} \cdot X_{t_{k}}^{\beta_1 1}) q_{k}^{3, q}
%= O_p(b_n^{-1/2})
%\end{eqnarray*}
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n-2} q_{k}^{2, p} q_{k+2}^{3, q}
%=\sum_{k=2}^{b_n-2} \frac{2}{3} (X_{t_{k-2}}^{\alpha_1 1} \cdot X_{t_{k-2}}^{\beta_1 1}) q_{k+2}^{3, q}
%= O_p(b_n^{-1/2})
%\end{eqnarray*}
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n-2} q_{k+2}^{2, p} q_{k+2}^{3, q}
%=\sum_{k=2}^{b_n-2} \frac{2}{3} (X_{t_{k}}^{\alpha_1 1} \cdot X_{t_{k}}^{\beta_1 1}) q_{k+2}^{3, q}
%= O_p(b_n^{-1/2})
%\end{eqnarray*}
%%
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n} q_{k}^{3, p} q_{k}^{2, q}
%=\sum_{k=2}^{b_n} q_{k}^{3, p} \frac{2}{3} (X_{t_{k-2}}^{\alpha_2 1} \cdot X_{t_{k-2}}^{\beta_2 1})
%= O_p(b_n^{-1/2})
%\end{eqnarray*}
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n-2} q_{k+2}^{3, p} q_{k}^{2, q}
%=\sum_{k=2}^{b_n-2} q_{k+2}^{3, p} \frac{2}{3} (X_{t_{k-2}}^{\alpha_2 1} \cdot X_{t_{k-2}}^{\beta_2 1})
%= O_p(b_n^{-1/2})
%\end{eqnarray*}
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n-2} q_{k}^{3, p} q_{k+2}^{2, q}
%=\sum_{k=2}^{b_n-2} q_{k}^{3, p} \frac{2}{3} (X_{t_{k}}^{\alpha_2 1} \cdot X_{t_{k}}^{\beta_2 1})
%= O_p(b_n^{-1/2})
%\end{eqnarray*}
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n-2} q_{k+2}^{3, p} q_{k+2}^{2, q}
%=\sum_{k=2}^{b_n-2} q_{k+2}^{3, p} \frac{2}{3} (X_{t_{k}}^{\alpha_2 1} \cdot X_{t_{k}}^{\beta_2 1})
%= O_p(b_n^{-1/2})
%\end{eqnarray*}
%
and
\begin{align} \label{3k+d3k}
\Big(\frac{T}{b_n} \Big)^{-1} \sum_{k=2}^{b_n-2} q_{k+2}^{3, p} q_{k}^{3, q}
= O_p(b_n^{-1/2}).
\end{align}
%
By Lemma 4 in \cite{Kimura2016} and simple calculus, it holds that
\begin{align} \label{3k3k}
&\Big(\frac{T}{b_n} \Big)^{-1} \sum_{k=2}^{b_n} q_{k}^{3, p} q_{k}^{3, q} \notag\\
&= \Big(\frac{T}{b_n} \Big)^{-1} \sum_{k=2}^{b_n}
\Big\{\sum_{i,j = 1}^{r}
(X^{\alpha_1 1}_{i, t_{k-2}} X^{\beta_1 1}_{j, t_{k-2}} + X^{\alpha_1 1}_{j, t_{k-2}} X^{\beta_1 1}_{i, t_{k-2}}) \notag\\
&\qquad\qquad\qquad\qquad\quad
\times
\Big(\int_{I_{k-1}} \int_{t_{k-2}}^{t} H_k(s) \ dw^{i}_s H_k(t) \ dw^{j}_t \notag\\
&\qquad\qquad\qquad\qquad\qquad
+ \int_{I_{k}} \int_{t_{k-1}}^{t} K_k(s) \ dw^{i}_s K_k(t) \ dw^{j}_t \notag\\
&\qquad\qquad\qquad\qquad\qquad
+ \int_{I_{k-1}} H_k(s) \ dw^{i}_s \int_{I_{k}} K_k(s) \ dw^{j}_s \Big) \Big\} \notag\\
&\qquad\qquad\qquad
\times
\Big\{ \sum_{i,j = 1}^{r}
(X^{\alpha_2 1}_{i, t_{k-2}} X^{\beta_2 1}_{j, t_{k-2}} + X^{\alpha_2 1}_{j, t_{k-2}} X^{\beta_2 1}_{i, t_{k-2}}) \notag\\
&\qquad\qquad\qquad\qquad\quad
\times
\Big(\int_{I_{k-1}} \int_{t_{k-2}}^{t} H_k(s) \ dw^{i}_s H_k(t) \ dw^{j}_t \notag\\
&\qquad\qquad\qquad\qquad\qquad
+ \int_{I_{k}} \int_{t_{k-1}}^{t} K_k(s) \ dw^{i}_s K_k(t) \ dw^{j}_t \notag\\
&\qquad\qquad\qquad\qquad\qquad
+ \int_{I_{k-1}} H_k(s) \ dw^{i}_s \int_{I_{k}} K_k(s) \ dw^{j}_s \Big) \Big\} \notag\\
&=
\Big(\frac{T}{b_n} \Big)^{-1} \sum_{k=2}^{b_n}
\Big\{\sum_{i,j = 1}^{r}
(X^{\alpha_1 1}_{i, t_{k-2}} X^{\beta_1 1}_{j, t_{k-2}} + X^{\alpha_1 1}_{j, t_{k-2}} X^{\beta_1 1}_{i, t_{k-2}}) \notag\\
&\qquad\qquad\qquad\qquad\quad
\times
(X^{\alpha_2 1}_{i, t_{k-2}} X^{\beta_2 1}_{j, t_{k-2}} + X^{\alpha_2 1}_{j, t_{k-2}} X^{\beta_2 1}_{i, t_{k-2}}) \Big\} \notag\\
&\qquad\qquad\qquad
\times
\Big\{ \int_0^T \int_0^t (\tilde{H}_k(s))^2 \ ds (\tilde{H}_k(t))^2 \ dt \notag\\
&\qquad\qquad\qquad\qquad
+ \int_0^T \int_0^t (\tilde{K}_k(s))^2 \ ds (\tilde{K}_k(t))^2 \ dt \notag\\
&\qquad\qquad\qquad\qquad
+ \int_0^T (\tilde{H}_k(s))^2 \ ds \int_0^T (\tilde{H}_k(s))^2 \ ds \Big\}
+ O_p(b_n^{-1/2}) \notag\\
&= \frac{4}{9} \sum_{k=2}^{b_n} \Big\{
(X^{\alpha_1 1}_{t_{k-2}} \cdot X^{\alpha_2 1}_{t_{k-2}})(X^{\beta_1 1}_{t_{k-2}} \cdot X^{\beta_2 1}_{t_{k-2}}) \notag\\
&\qquad\qquad
+ (X^{\alpha_1 1}_{t_{k-2}} \cdot X^{\beta_2 1}_{t_{k-2}})(X^{\beta_1 1}_{t_{k-2}} \cdot X^{\alpha_2 1}_{t_{k-2}}) \Big\}
\Big(\frac{T}{b_n}\Big)
+ O_p(b_n^{-1/2}),
\end{align}
where $\tilde{H}_k(s) = 1_{I_{k-1}}(s) H_k(s)$ and $\tilde{K}_k(s) = 1_{I_{k}}(s) K_k(s)$.
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n-2} q_{k+2}^{3, p} q_{k}^{3, q}
%= O_p(b_n^{-1/2}).
%\end{eqnarray*}
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n-2} q_{k}^{3, p} q_{k+2}^{3, q}
%= O_p(b_n^{-1/2}).
%\end{eqnarray*}
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n-2} q_{k+2}^{3, p} q_{k+2}^{3, q}
%&=& \frac{4}{9} \sum_{k=2}^{b_n-2} \{
%(X^{\alpha_1 1}_{t_{k}} \cdot X^{\alpha_2 1}_{t_{k}})(X^{\beta_1 1}_{t_{k}} \cdot X^{\beta_2 1}_{t_{k}})
%+ (X^{\alpha_1 1}_{t_{k}} \cdot X^{\beta_2 1}_{t_{k}})(X^{\beta_1 1}_{t_{k}} \cdot X^{\alpha_2 1}_{t_{k}}) \}
%\frac{T}{b_n} \\
%&&+ O_p(b_n^{-1/2}).
%\end{eqnarray*}
%
Therefore, we have
\begin{align*}
%\Big(\frac{T}{b_n} \Big)^{-1}
\mathcal{R}_{n,1}^{1,p q}
&= \frac{9}{8} \bigg\{ \frac{4}{9} \sum_{k=2}^{b_n} (X_{t_{k-2}}^{\alpha_1 1} \cdot X_{t_{k-2}}^{\beta_1 1})
(X_{t_{k-2}}^{\alpha_2 1} \cdot X_{t_{k-2}}^{\beta_2 1}) \Big(\frac{T}{b_n}\Big) \\
&\qquad
+ \frac{4}{9} \sum_{k=2}^{b_n} \Big\{
(X^{\alpha_1 1}_{t_{k-2}} \cdot X^{\alpha_2 1}_{t_{k-2}})(X^{\beta_1 1}_{t_{k-2}} \cdot X^{\beta_2 1}_{t_{k-2}}) \\
&\qquad\qquad\qquad
+ (X^{\alpha_1 1}_{t_{k-2}} \cdot X^{\beta_2 1}_{t_{k-2}})(X^{\beta_1 1}_{t_{k-2}} \cdot X^{\alpha_2 1}_{t_{k-2}}) \Big\}
(\frac{T}{b_n}\Big) \\
&\qquad
- \frac{1}{2} \Big\{ \frac{4}{9} \sum_{k=2}^{b_n-2} (X_{t_{k-2}}^{\alpha_1 1} \cdot X_{t_{k-2}}^{\beta_1 1})
(X_{t_{k}}^{\alpha_2 1} \cdot X_{t_{k}}^{\beta_2 1}) \Big(\frac{T}{b_n}\Big) \\
&\qquad\qquad
+ \frac{4}{9} \sum_{k=2}^{b_n-2} (X_{t_{k}}^{\alpha_1 1} \cdot X_{t_{k}}^{\beta_1 1})
(X_{t_{k-2}}^{\alpha_2 1} \cdot X_{t_{k-2}}^{\beta_2 1}) \Big(\frac{T}{b_n}\Big) \Big\} \bigg\} \\
&\qquad
+ O_p(b_n^{-1/2}) \\
&= \gamma^{p q} + o_p(b_n^{-1/2}) + O_p(b_n^{-1/2})
\end{align*}
and
\begin{align*}
%\Big(\frac{T}{b_n} \Big)^{-1}
\mathcal{R}_{n,2}^{1,p q}
&= \frac{9}{16} %\frac{9}{8} \frac{1}{2}
\bigg\{
\frac{4}{9} \sum_{k=2}^{b_n-2} (X_{t_{k}}^{\alpha_1 1} \cdot X_{t_{k}}^{\beta_1 1})
(X_{t_{k}}^{\alpha_2 1} \cdot X_{t_{k}}^{\beta_2 1}) \Big(\frac{T}{b_n}\Big) \\
&\qquad\quad
+ \frac{4}{9} \sum_{k=2}^{b_n-2} \Big\{
(X^{\alpha_1 1}_{t_{k-2}} \cdot X^{\alpha_2 1}_{t_{k-2}})(X^{\beta_1 1}_{t_{k-2}} \cdot X^{\beta_2 1}_{t_{k-2}}) \\
&\qquad\qquad\qquad\quad
+ (X^{\alpha_1 1}_{t_{k-2}} \cdot X^{\beta_2 1}_{t_{k-2}})(X^{\beta_1 1}_{t_{k-2}} \cdot X^{\alpha_2 1}_{t_{k-2}}) \Big\}
\Big(\frac{T}{b_n}\Big) \\
&\qquad\quad
- \frac{4}{9} \sum_{k=2}^{b_n-2} (X_{t_{k}}^{\alpha_1 1} \cdot X_{t_{k}}^{\beta_1 1})
(X_{t_{k-2}}^{\alpha_2 1} \cdot X_{t_{k-2}}^{\beta_2 1}) \Big(\frac{T}{b_n}\Big) \\
&\qquad\quad
- \frac{4}{9} \sum_{k=2}^{b_n-2} (X_{t_{k-2}}^{\alpha_1 1} \cdot X_{t_{k-2}}^{\beta_1 1})
(X_{t_{k}}^{\alpha_2 1} \cdot X_{t_{k}}^{\beta_2 1}) \Big(\frac{T}{b_n}\Big) \\
&\qquad\quad
+ \frac{4}{9} \sum_{k=2}^{b_n-2} (X_{t_{k-2}}^{\alpha_1 1} \cdot X_{t_{k-2}}^{\beta_1 1})
(X_{t_{k-2}}^{\alpha_2 1} \cdot X_{t_{k-2}}^{\beta_2 1}) \Big(\frac{T}{b_n}\Big) \\
&\qquad\quad
+ \frac{4}{9} \sum_{k=2}^{b_n-2} \Big\{
(X^{\alpha_1 1}_{t_{k-2}} \cdot X^{\alpha_2 1}_{t_{k-2}})(X^{\beta_1 1}_{t_{k-2}} \cdot X^{\beta_2 1}_{t_{k-2}}) \\
&\qquad\qquad\qquad\quad
+ (X^{\alpha_1 1}_{t_{k-2}} \cdot X^{\beta_2 1}_{t_{k-2}})(X^{\beta_1 1}_{t_{k-2}} \cdot X^{\alpha_2 1}_{t_{k-2}}) \Big\}
\Big(\frac{T}{b_n}\Big) \bigg\} \\
&\qquad
+ O_p(b_n^{-1/2}). \\
&= \gamma^{p q} + o_p(b_n^{-1/2}) + O_p(b_n^{-1/2}).
\end{align*}
The last $o_p(b_n^{-1/2})$s are derived from the proof of Lemma 3 in \cite{Kimura2016} and simple calculus.
%and
%\begin{eqnarray*}
%abcd - a'b'c'd' = (a-a')bcd + a'(b-b')cd + a'b'(c-c')d + a'b'c'(d-d').
%\end{eqnarray*}
\qed
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection*{Proof of Theorem \ref{kernel}}
%
%Let
In the similar way to the proof of Therem \ref{nonkernel}, let
\begin{align*}
G_{n,h}^{pq}
&= \frac{9}{8} \sum_{k=2}^{b_n} \Big\{
\partial_{h} \{ \tilde{\chi}^{n, \alpha_1} , \tilde{\chi}^{n, \alpha_2} \}_{k} \
\partial_{h} \{ \tilde{\chi}^{n, \beta_1} , \tilde{\chi}^{n, \beta_2} \}_{k} \\
&\qquad\qquad
+ \partial_{h} \{ \tilde{\chi}^{n, \alpha_1} , \tilde{\chi}^{n, \beta_2} \}_{k} \
\partial_{h} \{ \tilde{\chi}^{n, \beta_1} , \tilde{\chi}^{n, \alpha_2} \}_{k} \Big\} \Big(\frac{T}{b_n} \Big).
\end{align*}
%
%$\bm{\chi}^n = (\chi^{n, 1}, \chi^{n, 2})$
%
%$\chi^{n, \alpha}_t = \int_{0}^{t} a_n X^{\alpha}_{s} \ ds$
%
%For a stochastic process $V$,
%\begin{eqnarray*}
%\tilde{V}_{k} := \frac{\Delta_{k}V}{a_n h_{k}} = \frac{\Delta_{k}V}{a_n \delta_n}.
%\end{eqnarray*}
%
%
\begin{lemma*} \label{lemma1h}
(a)
$%(T/b_n)^{-1}
\hat{\gamma}^{p q}_{n,h} = %(T/b_n)^{-1}
G_{n,h}^{p q} + o_p(1)$
under any one of [B], [B$'$] and [B$^{\sharp^{\sharp}}$].
(b)
$%(T/b_n)^{-1}
\hat{\gamma}^{p q}_{n,h} = %(T/b_n)^{-1}
G_{n,h}^{p q} + o_p(b_n^{-1/2})$
under [B$^{\sharp^{\sharp}}$].
\end{lemma*}
\proof
%
%$s_{k} = (s_{k}^{12}, s_{k}^{21}, s_{k}^{11}, s_{k}^{22})^{\star}$,
%$t_{k} = (t_{k}^{12}, t_{k}^{21}, t_{k}^{11}, t_{k}^{22})^{\star}$.
%
We write
\begin{align*}
\check{s}_{k, h}%^{\alpha \beta}
:= \sum_{l=(k - n(h) +1) \vee 2}^{k} s_{l}%^{\alpha \beta}
\ h^{-1},
\qquad
\check{t}_{k, h}%^{\alpha \beta}
:= \sum_{l=(k - n(h) +1) \vee 2}^{k} t_{l}%^{\alpha \beta}
\ h^{-1},
\end{align*}
%$\check{s}_{k, h} = (\check{s}_{k, h}^{12}, \check{s}_{k, h}^{11}, \check{s}_{k, h}^{22})^{\star}$,
%$\check{t}_{k, h} = (\check{t}_{k, h}^{12}, \check{t}_{k, h}^{11}, \check{t}_{k, h}^{22})^{\star}$.
$\hat{\gamma}_{n,h} = (\hat{\gamma}^{p q}_{n,h})_{p, q = (1,2),(2,1),(1,1),(2,2)}$, and
${G}_{n,h} = ({G}^{p q}_{n,h})_{p, q = (1,2),(2,1),(1,1),(2,2)}$.
%
It holds that
\begin{align*}
&b_n^{\rho} \Big(\frac{9}{8} \Big)^{-1} %\Big(\frac{T}{b_n} \Big)^{-1}
| \hat{\gamma}_{n,h} - G_{n,h} | \\
&= b_n^{\rho} 2 %\Big(\frac{T}{b_n} \Big)^{-1}
\bigg| \sum_{k=2}^{b_n} (\check{s}_{k, h} \otimes \check{s}_{k, h} - \check{t}_{k, h} \otimes \check{t}_{k, h}) \bigg| \Big(\frac{T}{b_n} \Big) \\
&= b_n^{\rho} 2 %\Big(\frac{T}{b_n} \Big)^{-1}
\bigg| \sum_{k=2}^{b_n} (\check{s}_{k, h} - \check{t}_{k, h}) \otimes (\check{s}_{k, h} - \check{t}_{k, h}) \\
&\qquad\qquad\qquad\qquad
+ (\check{s}_{k, h} - \check{t}_{k, h}) \otimes \check{t}_{k, h}
+ \check{t}_{k, h} \otimes (\check{s}_{k, h} - \check{t}_{k, h}) \bigg| \Big(\frac{T}{b_n} \Big) \\
&\leq b_n^{\rho} 2 %\Big(\frac{T}{b_n} \Big)^{-1}
\sum_{k=2}^{b_n}
\Big\{ \Big| (\check{s}_{k, h} - \check{t}_{k, h}) \otimes (\check{s}_{k, h} - \check{t}_{k, h}) \Big| \\
&\qquad\qquad\qquad\qquad
+ \Big| (\check{s}_{k, h} - \check{t}_{k, h}) \otimes \check{t}_{k, h} \Big|
+ \Big| \check{t}_{k, h} \otimes (\check{s}_{k, h} - \check{t}_{k, h}) \Big| \Big\}
\Big(\frac{T}{b_n} \Big) \\
&= b_n^{\rho} 2 %\Big(\frac{T}{b_n} \Big)^{-1}
\sum_{k=2}^{b_n} | \check{s}_{k, h} - \check{t}_{k, h} |^2 \Big(\frac{T}{b_n} \Big)
+ b_n^{\rho} 4 %\Big(\frac{T}{b_n} \Big)^{-1}
\sum_{k=2}^{b_n} | \check{t}_{k, h} | | \check{s}_{k, h} - \check{t}_{k, h} | \Big(\frac{T}{b_n}\Big) \\
&\leq 2 \bigg\{ b_n^{\rho/2} \Big(\frac{T}{b_n} \Big)^{-1/2} \sum_{k=2}^{b_n} | \check{s}_{k, h} - \check{t}_{k, h} | \Big(\frac{T}{b_n} \Big) \bigg\}^2 \\
&\quad
+ 4 \Big\{ \max_k |\check{t}_{k, h} | \Big\}
\bigg\{ b_n^{\rho} \sum_{k=2}^{b_n} | \check{s}_{k, h} - \check{t}_{k, h} | \Big(\frac{T}{b_n} \Big) \bigg\}.
\end{align*}
We obtain the conclusion from the inequalities
\begin{align*}
\sum_{k=2}^{b_n} | \check{s}_{k, h} - \check{t}_{k, h} | \Big(\frac{T}{b_n} \Big)
\leq \sum_{k=2}^{b_n} \sum_{l=(k - n(h) +1) \vee 2}^{k} | s_{l} - t_{l} | \ h^{-1} \Big(\frac{T}{b_n} \Big)
\leq \sum_{k=2}^{b_n} | s_{k} - t_{k} |,
\end{align*}
%\begin{eqnarray*}
%\max_{k} | (\frac{T}{b_n}) \check{t}_{k, h} |
%\leq \max_{k} | t_{k} |,
%\end{eqnarray*}
%\begin{eqnarray*}
%|S_n - T_n|
%\leq \sum_{k=2}^{b_n} | s_{k} - t_{k} |
%\leq \sum_{k=2}^{b_n} | \tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1}|^2
%+ 2 \{ \sum_{k=2}^{b_n} | \tilde{\mathbb{M}}^n_{k} - \tilde{\mathbb{M}}^n_{k-1}|^2 \}^{1/2}
%\{ \sum_{k=2}^{b_n} | \tilde{\bm{\chi}}^n_{k} - \tilde{\bm{\chi}}^n_{k-1}|^2 \}^{1/2},
%\end{eqnarray*}
%and
\begin{align*}
\max_{k} | \check{t}_{k, h} | \leq \max_{k} | t_{k} |,
\end{align*}
and the proof of Lemma \ref{lemma1i}.
\qed
%
%Under [B$^{\sharp^{\sharp}}$], we can prove
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} \hat{\gamma}^{p q}_n = (\frac{T}{b_n})^{-1} G^{p q}_{n, i} + o_p(b_n^{-1/2}).
%\end{eqnarray*}
%by similar argument after the proof of Lemma 4.1.
%
%cf.
%\begin{eqnarray*}
%\sqrt{b_n} (\frac{9}{8})^{-1} (\frac{T}{b_n})^{-1} | \hat{\gamma}_{n,h} - G_{n,h} |
%&\leq& 2 \{ (b_n)^{-1/4} (\frac{T}{b_n})^{-1/2} \sum_{k=2}^{b_n} | \check{s}_{k, h} - \check{t}_{k, h} | (\frac{T}{b_n}) \}^2 \\
%&&+ 4 \{ \max_k |\check{t}_{k, h} | \}
%\{\sqrt{b_n} \sum_{k=2}^{b_n} | \check{s}_{k, h} - \check{t}_{k, h} | (\frac{T}{b_n})\}
%\end{eqnarray*}
%
\begin{proposition*} \label{prop1h}
$%(T/b_n)^{-1}
\hat{\gamma}^{p q}_{n,h} = %(T/b_n)^{-1}
G_{n,h}^{p q} + o_p(b_n^{-1/2})$
under [B] or [B'], and [C].
\end{proposition*}
%
\proof
The conclusion is derived from the proof of Lemma \ref{lemma1h}, that of Proposition \ref{prop1i}, and the inequality
\begin{align*}
b_n^{\rho} %\Big(\frac{T}{b_n} \Big)^{-1}
\sum_{k=2}^{b_n} | \check{s}_{k, h} - \check{t}_{k, h} |^2 \Big(\frac{T}{b_n} \Big)
\leq
b_n^{\rho} \sum_{k=2}^{b_n} | {s}_{k} - {t}_{k} |^2 \Big(\frac{T}{b_n} \Big)^{-1}.
\end{align*}
\qed
%
\begin{lemma*} \label{lemma2h}
$b_n^{\rho} \sum_{l=(k - n(h) +1) \vee 2}^{k} F_{l} G_{l} h^{-1} \to^p 0$
for all pairs of \\
$(F_{l}^, G_{l}) \in \{ A_{k}^{\alpha}, B_{k}^{\alpha}, C_{k}^{\alpha}, D_{k}^{\alpha} ; \alpha=1, 2 \}^2 \backslash \{(A_{l}^{\alpha}, A_{l}^{\beta}) ; \alpha, \beta =1, 2 \}$.
\end{lemma*}
The proof of this lemma is very similar to proof of Lemma 2 in \cite{Kimura2016}.
\proof
Let
$\mathcal{Q}_{n, h} = b_n^{\rho} \sum_{l=(k - n(h) +1) \vee 2}^{k} F_{l} G_{l} h^{-1}$.
%
%For $F_{l} = A_{l}^{\alpha}$ and $G_{l} = C_{l}^{\beta}$,
%$E[|\mathcal{Q}_{n, h}|] = O(n(h) h^{-1} b_n^{-3/2}) = O(b_n^{-1/2})$.
%
For $F_{l} = A_{l}^{\alpha}$ and $G_{l} = C_{l}^{\beta}$,
$E[(|\mathcal{Q}_{n, h}|)^2] = O(b_n^{2 \rho} n(h) h^{-2} b_n^{-3}) = O(h^{-1} b_n^{2\rho -2})=o(1)$.
%
For $F_{l} = A_{l}^{\alpha}$ and $G_{l} = D_{l}^{\beta}$,
$E[|\mathcal{Q}_{n, h}|] = O( b_n^{\rho} n(h) h^{-1} b_n^{-3/2}) o(1) = o(b_n^{\rho -1/2})=o(1)$.
%
%
%For $F_{l} = A_{l}^{\alpha}$ and $B_{l} = C_{l}^{\beta}$,
%$E[|\mathcal{Q}_{n, h}|^2] = O(n(h) h^{-2} b_n^{-3}) = O(b_n^{-2} h^{-1})$.
%Therefore, $\mathcal{Q}_{n, h} = O_p(b_n^{-1/2} (h^{-1} b_n^{-1/2})) = O_p(b_n^{-1/2})$.
%
%
%For $F_{l} = A_{l}^{\alpha}$ and $G_{l} = B_{l}^{\beta}$,
%$E[|\mathcal{Q}_{n, h}|] = O(n(h) h^{-1} b_n^{-1}) o(1) = o(1)$.
%
For $F_{l} = A_{l}^{\alpha}$ and $G_{l} = B_{l}^{\beta}$,
$E[( |\mathcal{Q}_{n, h}|)^2] = O(b_n^{2 \rho} n(h) h^{-2} b_n^{-3}) = O(h^{-1} b_n^{2 \rho -2})=o(1)$.
\qed
%
%Actually, we can prove
%$\sum_{l=(k - n(h) +1)_{+}}^{k} A_{l}^{\alpha} B_{l}^{\beta} h^{-1} = o_p(b_n^{-1/2})$.
%
%For $F_{l} = A_{l}^{\alpha}$ and $G_{l} = C_{l}^{\beta}$,
%$E[(b_n^{1/2} |\mathcal{Q}_{n, h}|)^2] = O(b_n n(h) h^{-2} b_n^{-3}) = O(h^{-1} b_n^{-1})$.
%
%For $F_{l} = A_{l}^{\alpha}$ and $G_{l} = B_{l}^{\beta}$,
%$E[(b_n^{1/2} |\mathcal{Q}_{n, h}|)^2] = O(b_n n(h) h^{-2} b_n^{-3}) = O(h^{-1} b_n^{-1})$.
%
Now, let
\begin{align*}
Q_{k, h}^{1, \alpha \beta} = \sum_{l=(k - n(h) +1) \vee 2}^{k} A_{l}^{\alpha} A_{l}^{\beta} h^{-1}.
\end{align*}
By Lemma \ref{lemma2h} %4.3.
and (4)
(esp. $Q_{k, h}^{1, \alpha \beta} = O_p(1)$),
%we have
it holds that
\begin{align*}
%\Big(\frac{T}{b_n} \Big)^{-1}
G_{n,h}^{p q}
&= %\Big(\frac{T}{b_n} \Big)^{-1}
\frac{9}{8} \sum_{k=2}^{b_n} (Q_{k, h}^{1, \alpha_1 \alpha_2} Q_{k, h}^{1, \beta_1 \beta_2}
+ Q_{k, h}^{1, \alpha_1 \beta_2} Q_{k, h}^{1, \beta_1 \alpha_2}) \Big(\frac{T}{b_n} \Big) + o_p(b_n^{-1/2}) \\
&=: %\Big(\frac{T}{b_n} \Big)^{-1}
\mathcal{R}_{n,h}^{1, p q} + o_p(b_n^{-1/2}).
\end{align*}
%We can also prove
%\begin{eqnarray*}
%(\frac{T}{b_n})^{-1} G_{n,h}^{p q}
%= (\frac{T}{b_n})^{-1} \sum_{k=2}^{b_n} \frac{9}{8} (Q_{k, h}^{1, \alpha_1 \alpha_2} Q_{k, h}^{1, \beta_1 \beta_2}
%+ Q_{k, h}^{1, \alpha_1 \beta_2} Q_{k, h}^{1, \beta_1 \alpha_2}) (\frac{T}{b_n})^{2} + o_p(b_n^{-1/2})
%=: b_n \mathcal{R}_{n,h}^{1, p q} + o_p(b_n^{-1/2}).
%\end{eqnarray*}
%
By recombination of the multiple summation,
\begin{align*}
\mathcal{R}_{n,h}^{1, p q}
&= \frac{9}{8} \sum_{l=2}^{b_n - n(h)} \Big\{
n(h)
(q_{l}^{1, \alpha_1 \alpha_2} q_{l}^{1, \beta_1 \beta_2}
+ q_{l}^{1, \alpha_1 \beta_2} q_{l}^{1, \beta_1 \alpha_2}) \\
&\qquad\qquad\quad
+ (n(h) - 1)
(q_{l}^{1, \alpha_1 \alpha_2} q_{l+1}^{1, \beta_1 \beta_2}
+ q_{l}^{1, \alpha_1 \beta_2} q_{l+1}^{1, \beta_1 \alpha_2} \\
&\qquad\qquad\qquad\qquad\qquad
+ q_{l+1}^{1, \alpha_1 \alpha_2} q_{l}^{1, \beta_1 \beta_2}
+ q_{l+1}^{1, \alpha_1 \beta_2} q_{l}^{1, \beta_1 \alpha_2}) \\
&\qquad\qquad\quad
+ \sum_{k=2}^{n(h)-1} (n(h) - k)
(q_{l}^{1, \alpha_1 \alpha_2} q_{l+k}^{1, \beta_1 \beta_2}
+ q_{l}^{1, \alpha_1 \beta_2} q_{l+k}^{1, \beta_1 \alpha_2} \\
&\qquad\qquad\qquad\qquad\qquad\qquad\quad
+ q_{l+k}^{1, \alpha_1 \alpha_2} q_{l}^{1, \beta_1 \beta_2}
+ q_{l+k}^{1, \alpha_1 \beta_2} q_{l}^{1, \beta_1 \alpha_2})\Big\} h_n^{-2} \Big(\frac{T}{b_n} \Big) \\
&\quad
+ \frac{9}{8} \sum_{m=b_n - n(h) + 1}^{b_n} \Big\{
\Big( \sum_{l = b_n - n(h) + 1}^{m} q_{l}^{1, \alpha_1 \alpha_2} \Big)
\Big( \sum_{l = b_n - n(h) + 1}^{m} q_{l}^{1, \beta_1 \beta_2} \Big) \\
&\qquad\qquad\qquad\qquad
+
\Big( \sum_{l = b_n - n(h) + 1}^{m} q_{l}^{1, \alpha_1 \beta_2} \Big)
\Big( \sum_{l = b_n - n(h) + 1}^{m} q_{l}^{1, \beta_1 \alpha_2} \Big) \Big\} h_n^{-2} \Big(\frac{T}{b_n} \Big) \\
&=: \mathcal{B}_{n,h}^{1, p q} + \mathcal{T}_{n,h}^{1, p q},
\end{align*}
where $q_{l}^{1, \alpha \beta} = A_{l}^{\alpha} A_{l}^{\beta} = q_{l}^{2, \alpha \beta} + q_{l}^{3, \alpha \beta}$.
By (\ref{2k3k}), (\ref{3k+d3k}), (\ref{3k3k}) and similar evaluation, we have
$\mathcal{B}_{n,h}^{1, p q} = \mathcal{B}_{n,h}^{2, p q} + O_p(b_n^{-1/2})$,
where $\mathcal{B}_{n,h}^{2, p q}$ is defined in the same way of $\mathcal{B}_{n,h}^{1, p q}$
with $q_{l}^{2, \alpha \beta}$ instead of $q_{l}^{1, \alpha \beta}$.
It is easy to see that
$\mathcal{T}_{n,h}^{1, p q} = O_p(h_n)$.
%By recombination of the multiple summation,
By simple calculus,
we have
\begin{align*}
\mathcal{B}_{n,h}^{2, p q}
&=
\frac{9}{8} \sum_{l=2}^{b_n - n(h)} \sum_{k=1}^{n(h) -1} (n(h) - k) \\
&\qquad
\times \Big\{
(q_{l}^{2, \alpha_1 \alpha_2} (q_{l+k}^{2, \beta_1 \beta_2} - q_{l}^{2, \beta_1 \beta_2})
+ q_{l}^{2, \alpha_1 \beta_2} (q_{l+k}^{2, \beta_1 \alpha_2} - q_{l}^{2, \beta_1 \alpha_2}) \\
&\qquad\quad
+ (q_{l+k}^{2, \alpha_1 \alpha_2} - q_{l}^{2, \alpha_1 \alpha_2}) q_{l}^{2, \beta_1 \beta_2}
+ (q_{l+k}^{2, \alpha_1 \beta_2} - q_{l}^{2, \alpha_1 \beta_2}) q_{l}^{2, \beta_1 \alpha_2})
\Big\} h_n^{-2} \Big(\frac{T}{b_n} \Big) \\
&\quad
+ \frac{9}{8} \sum_{l=2}^{b_n} n(h)^2 %\Big\{
(q_{l}^{2, \alpha_1 \alpha_2} q_{l}^{2, \beta_1 \beta_2}
+ q_{l}^{2, \alpha_1 \beta_2} q_{l}^{2, \beta_1 \alpha_2}) %\Big\}
h_n^{-2} \Big(\frac{T}{b_n} \Big) \\
&\quad
- \frac{9}{8} \sum_{l=b_n - n(h)+1}^{b_n} n(h)^2 %\Big\{
(q_{l}^{2, \alpha_1 \alpha_2} q_{l}^{2, \beta_1 \beta_2}
+ q_{l}^{2, \alpha_1 \beta_2} q_{l}^{2, \beta_1 \alpha_2}) %\Big\}
h_n^{-2} \Big(\frac{T}{b_n} \Big) \\
&=: \mathcal{P}_{n,h}^{2, p q}
+ \mathcal{C}_{n,h}^{2, p q}
- \mathcal{T}_{n,h}^{2, p q}.
\end{align*}
%It is easy to see that
From the proof of Lemma 3 in \cite{Kimura2016} and simple calculus, it is derived that
$\mathcal{C}_{n,h}^{2, p q} = \gamma^{p q} + o_p(b_n^{-1/2})$.
%and
It is easy to see that
$\mathcal{T}_{n,h}^{2, p q} = o_p(h_n)$.
%
%The last $o_p(b_n^{-1/2})$s are derived from the proof of Lemma 3 in \cite{Kimura2016} and simple calculus.
%
Furthermore,
%By recombination of the multiple summation,
we have
\begin{align*}
\mathcal{P}_{n,h}^{2, p q}
&=
\frac{9}{8} \sum_{l=2}^{b_n - n(h)} \sum_{k=1}^{n(h) -1} (n(h) - k) \\
&\qquad
\times \Big\{
(q_{l}^{2, \alpha_1 \alpha_2} \sum_{m=1}^{k} (q_{l+m}^{2, \beta_1 \beta_2} - q_{l+m-1}^{2, \beta_1 \beta_2})
+ q_{l}^{2, \alpha_1 \beta_2} \sum_{m=1}^{k} (q_{l+m}^{2, \beta_1 \alpha_2} - q_{l+m-1}^{2, \beta_1 \alpha_2}) \\
&\qquad\quad
+ \sum_{m=1}^{k} (q_{l+m}^{2, \alpha_1 \alpha_2} - q_{l+m-1}^{2, \alpha_1 \alpha_2}) q_{l}^{2, \beta_1 \beta_2}
+ \sum_{m=1}^{k} (q_{l+m}^{2, \alpha_1 \beta_2} - q_{l+m-1}^{2, \alpha_1 \beta_2}) q_{l}^{2, \beta_1 \alpha_2})
\Big\} h_n^{-2} \Big(\frac{T}{b_n} \Big) \\
&=
\frac{9}{8} \sum_{l=2}^{b_n - n(h)} \sum_{k=1}^{n(h) -1} \frac{(n(h) - k)(n(h) - k+1)}{2} \\
&\qquad
\times \Big\{
(q_{l}^{2, \alpha_1 \alpha_2} (q_{l+k}^{2, \beta_1 \beta_2} - q_{l+k-1}^{2, \beta_1 \beta_2})
+ q_{l}^{2, \alpha_1 \beta_2} (q_{l+k}^{2, \beta_1 \alpha_2} - q_{l+k-1}^{2, \beta_1 \alpha_2}) \\
&\qquad\quad
+ (q_{l+k}^{2, \alpha_1 \alpha_2} - q_{l+k-1}^{2, \alpha_1 \alpha_2}) q_{l}^{2, \beta_1 \beta_2}
+ (q_{l+k}^{2, \alpha_1 \beta_2} - q_{l+k-1}^{2, \alpha_1 \beta_2}) q_{l}^{2, \beta_1 \alpha_2})
\Big\} h_n^{-2} \Big(\frac{T}{b_n} \Big) \\
&=
\frac{9}{8} \sum_{l=2}^{n(h)} \sum_{k=1}^{n(h) - l} \frac{(n(h) - k)(n(h) - k+1)}{2} \\
&\qquad
\times \Big\{
(q_{l}^{2, \alpha_1 \alpha_2} (q_{l+k}^{2, \beta_1 \beta_2} - q_{l+k-1}^{2, \beta_1 \beta_2})
+ q_{l}^{2, \alpha_1 \beta_2} (q_{l+k}^{2, \beta_1 \alpha_2} - q_{l+k-1}^{2, \beta_1 \alpha_2}) \\
&\qquad\quad
+ (q_{l+k}^{2, \alpha_1 \alpha_2} - q_{l+k-1}^{2, \alpha_1 \alpha_2}) q_{l}^{2, \beta_1 \beta_2}
+ (q_{l+k}^{2, \alpha_1 \beta_2} - q_{l+k-1}^{2, \alpha_1 \beta_2}) q_{l}^{2, \beta_1 \alpha_2})
\Big\} h_n^{-2} \Big(\frac{T}{b_n} \Big) \\
&\quad
+
\frac{9}{8} \sum_{l'=n(h)+1}^{b_n} \sum_{k=1}^{n(h) -1} \frac{(n(h) - k)(n(h) - k+1)}{2} \\
&\qquad
\times \Big\{
(q_{l'-k}^{2, \alpha_1 \alpha_2} (q_{l'}^{2, \beta_1 \beta_2} - q_{l'-1}^{2, \beta_1 \beta_2})
+ q_{l'-k}^{2, \alpha_1 \beta_2} (q_{l'}^{2, \beta_1 \alpha_2} - q_{l'-1}^{2, \beta_1 \alpha_2}) \\
&\qquad\quad
+ (q_{l'}^{2, \alpha_1 \alpha_2} - q_{l'-1}^{2, \alpha_1 \alpha_2}) q_{l'-k}^{2, \beta_1 \beta_2}
+ (q_{l'}^{2, \alpha_1 \beta_2} - q_{l'-1}^{2, \alpha_1 \beta_2}) q_{l'-k}^{2, \beta_1 \alpha_2})
\Big\} h_n^{-2} \Big(\frac{T}{b_n} \Big) \\
&\quad
-
\frac{9}{8} \sum_{l=b_n - n(h) + 1}^{b_n - 1} \sum_{k=1}^{b_n - l} \frac{(n(h) - k)(n(h) - k+1)}{2} \\
&\qquad
\times \Big\{
(q_{l}^{2, \alpha_1 \alpha_2} (q_{l+k}^{2, \beta_1 \beta_2} - q_{l+k-1}^{2, \beta_1 \beta_2})
+ q_{l}^{2, \alpha_1 \beta_2} (q_{l+k}^{2, \beta_1 \alpha_2} - q_{l+k-1}^{2, \beta_1 \alpha_2}) \\
&\qquad\quad
+ (q_{l+k}^{2, \alpha_1 \alpha_2} - q_{l+k-1}^{2, \alpha_1 \alpha_2}) q_{l}^{2, \beta_1 \beta_2}
+ (q_{l+k}^{2, \alpha_1 \beta_2} - q_{l+k-1}^{2, \alpha_1 \beta_2}) q_{l}^{2, \beta_1 \alpha_2})
\Big\} h_n^{-2} \Big(\frac{T}{b_n} \Big) \\
&=:
\mathcal{H}_{n,h}^{3, p q}
+ \mathcal{P}_{n,h}^{3, p q}
- \mathcal{T}_{n,h}^{3, p q}.
\end{align*}
%We can see directly that
%$\mathcal{H}_{n,h}^{3, p q} = O_p(h_n^2 b_n^{1/2})$
%and
%$\mathcal{T}_{n,h}^{3, p q} = O_p(h_n^2 b_n^{1/2})$.
%Furthermore, by recombination of the multiple summation, we also have
$\mathcal{H}_{n,h}^{3, p q} = O_p(h_n)$
and
$\mathcal{T}_{n,h}^{3, p q} = O_p(h_n)$
are derived from the following calculus.
\begin{align*}
&\frac{(n(h) - k)(n(h) - k+1)}{2} (q_{l+k}^{2, \beta_1 \beta_2} - q_{l+k-1}^{2, \beta_1 \beta_2})\\
&= \frac{(n(h) - k)(n(h) - (k-1))}{2} q_{l+k}^{2, \beta_1 \beta_2}
- \frac{(n(h) - (k-1))(n(h) - (k-2))}{2} q_{l+(k-1)}^{2, \beta_1 \beta_2} \\
&\quad+ (n(h) - (k-1))q_{l+(k-1)}^{2, \beta_1 \beta_2}.
\end{align*}
Here, we set
\begin{align*}
\mathcal{P}_{n,h}^{3, p q}
= \sum_{l'=n(h)+1}^{b_n} p_{n,h, l'}^{3, p q}.
\end{align*}
and we have
\begin{align*}
\sum_{l'=n(h)+1}^{b_n} E\Big[p_{n,h, l'}^{3, p q} \Big| \mathcal{F}_{l'-3}^{n}\Big] &= O_p(h_n), \\
\sum_{l'=n(h)+1}^{b_n} E\Big[ \big|p_{n,h, l'}^{3, p q} \big|^2 \Big| \mathcal{F}_{l'-3}^{n}\Big] &= O_p(h_n^2)
\end{align*}
and $\{p_{n,h, l'}^{3, p q}\}_{l'=2}^{b_n}$ is $\{\mathcal{F}_{l'-2}^{n}\}_{l'=2}^{b_n}$-adapted.
Therefore, by Lenglart's inequality and its application, it holds that
$\mathcal{P}_{n,h}^{3, p q} = O_p(h_n)$.
Here, we proved that $G^{pq}_{n,h} = \gamma^{p q} + o_p(b_n^{-1/2}) + O_p(b_n^{-1/2}) + O_p(h_n)$.
%We reached the conclusion if $h_n = O(b_n^{-1/2})$ and $h_n^{-1} b_n^{-1} \to 0$ as $n \to \infty$.
We reached the conclusion.
\qed
\section{Simulation} \label{simulation}
We consider the following example.
$\mathbb{X}$ is an $\mathbb{R}_{+}^2$-valued It\^o process satisfying
\begin{eqnarray*}
X^{1}_t
&=& X^{1}_{0} + \int_{0}^{t} X^{1}_{s} \mu_{1} \ ds + \int_{0}^{t} X^{1}_{s} \sigma_{1} \ dw^{1}_{s} \\
X^{2}_t
&=& X^{2}_{0} + \int_{0}^{t} X^{2}_{s} \mu_{2} \ ds + \int_{0}^{t} X^{2}_{s} \rho \sigma_{2} \ dw^{1}_{s}
+ \int_{0}^{t} X^{2}_{s} \sqrt{1 - \rho^2} \sigma_{2} \ dw^{2}_{s},
\end{eqnarray*}
where $w^1$ and $w^2$ are independent Wiener processes, and $t \in [0, T]$.
$\mathbb{Y}$ is a counting process with intensity process $a_n \mathbb{X}$.
In this setting, [C] is satisfied.
In various $a_n$, $b_n$ settings,
We examine the asymptotic variance estimators
$\hat{\Xi}_{n,*}$, for $* = 1, 2, w, m, n$, where
$\hat{\Xi}_{n,w} = \hat{\Xi}_{n,h}$, $h = h_n = T b_n^{-0.25}$;
$\hat{\Xi}_{n,m} = \hat{\Xi}_{n,h}$, $h = h_n = T b_n^{-0.5}$;
$\hat{\Xi}_{n,n} = \hat{\Xi}_{n,h}$, $h = h_n = T b_n^{-0.75}$.
To reflect the condition that $\lim_{n \to \infty} b_n^r / a_n = 0$ ($r = 2, 2.5, 3$),
we see the case that $a_n = b_n^r$ ($r = 2, 2.5, 3, 3.5$).
We calculate these with 1000 paths and compare these by mean squared error (MSE).
Now, we fix the parameters
$\mu_1 = 0.2,
\mu_2 = 0.3,
\sigma_1 = 0.2,
\sigma_2 = 0.3,
\rho = 0.7,
X^1_0 = 1,
X^2_0 = 2,
T = 1$.
% latex table generated in R 3.4.3 by xtable 1.8-2 package
% Sun Jun 24 16:11:12 2018
\begin{table}[tbp]
\centering
\caption{MSE of asymptotic variance estimators; $a_n = b_n^2$}
\begin{tabular}{rrrrrrrr}
\hline
$*$ & $b_n = 2^4$ & $b_n = 2^5$ & $b_n = 2^6$ & $b_n = 2^7$ & $b_n = 2^8$ & $b_n = 2^9$ & $b_n = 2^{10}$ \\
\hline
1 & 0.6514 & 0.6775 & 0.6931 & 0.6835 & 0.6892 & 0.6923 & 0.6908 \\
2 & 0.4758 & 0.5838 & 0.6427 & 0.6537 & 0.6749 & 0.6845 & 0.6865 \\
w & 0.2102 & 0.2279 & 0.2783 & 0.3554 & 0.3980 & 0.4213 & 0.4667 \\
m & 0.6063 & 0.3973 & 0.7440 & 0.6470 & 0.7390 & 0.6298 & 0.7188 \\
n & 1.3349 & 0.7355 & 0.3196 & 0.8940 & 1.3771 & 0.5679 & 0.6630 \\
\hline
\end{tabular} \label{MSE2.0}
\end{table}
% latex table generated in R 3.4.3 by xtable 1.8-2 package
% Sun Jun 24 16:56:39 2018
\begin{table}[tbp]
\centering
\caption{MSE of asymptotic variance estimators; $a_n = b_n^{2.5}$}
\begin{tabular}{rrrrrrrr}
\hline
$*$ & $b_n = 2^4$ & $b_n = 2^5$ & $b_n = 2^6$ & $b_n = 2^7$ & $b_n = 2^8$ & $b_n = 2^9$ & $b_n = 2^{10}$ \\
\hline
1 & 0.6634 & 0.6204 & 0.6516 & 0.6101 & 0.5332 & 0.4594 & 0.3712 \\
2 & 0.4831 & 0.5289 & 0.5987 & 0.5828 & 0.5212 & 0.4540 & 0.3690 \\
w & 0.2066 & 0.2113 & 0.2463 & 0.2918 & 0.2918 & 0.2617 & 0.2317 \\
m & 0.5864 & 0.3543 & 0.6548 & 0.5325 & 0.5496 & 0.4013 & 0.3769 \\
n & 1.2788 & 0.6335 & 0.2680 & 0.7180 & 0.9890 & 0.3297 & 0.3145 \\
\hline
\end{tabular} \label{MSE2.5}
\end{table}
% latex table generated in R 3.4.3 by xtable 1.8-2 package
% Sun Jun 24 16:57:47 2018
\begin{table}[tbp]
\centering
\caption{MSE of asymptotic variance estimators; $a_n = b_n^{3}$}
\begin{tabular}{rrrrrrrr}
\hline
$*$ & $b_n = 2^4$ & $b_n = 2^5$ & $b_n = 2^6$ & $b_n = 2^7$ & $b_n = 2^8$ & $b_n = 2^9$ & $b_n = 2^{10}$ \\
\hline
1 & 0.5676 & 0.4363 & 0.2485 & 0.1052 & 0.0364 & 0.0113 & 0.0038 \\
2 & 0.4185 & 0.3663 & 0.2259 & 0.0994 & 0.0351 & 0.0110 & 0.0037 \\
w & 0.1854 & 0.1289 & 0.0736 & 0.0344 & 0.0106 & 0.0025 & 0.0013 \\
m & 0.5149 & 0.2247 & 0.2333 & 0.0798 & 0.0348 & 0.0069 & 0.0034 \\
n & 1.1188 & 0.4379 & 0.0818 & 0.1119 & 0.0885 & 0.0037 & 0.0018 \\
\hline
\end{tabular} \label{MSE3.0}
\end{table}
% latex table generated in R 3.4.3 by xtable 1.8-2 package
% Sun Jun 24 16:58:55 2018
\begin{table}[tbp]
\centering
\caption{MSE of asymptotic variance estimators; $a_n = b_n^{3.5}$}
\begin{tabular}{rrrrrrrr}
\hline
$*$ & $b_n = 2^4$ & $b_n = 2^5$ & $b_n = 2^6$ & $b_n = 2^7$ & $b_n = 2^8$ & $b_n = 2^9$ & $b_n = 2^{10}$ \\
\hline
1 & 0.3284 & 0.1144 & 0.0423 & 0.0135 & 0.0058 & 0.0027 & 0.0015 \\
2 & 0.2468 & 0.0948 & 0.0391 & 0.0128 & 0.0057 & 0.0027 & 0.0015 \\
w & 0.1003 & 0.0288 & 0.0132 & 0.0082 & 0.0059 & 0.0043 & 0.0027 \\
m & 0.2784 & 0.0471 & 0.0313 & 0.0086 & 0.0041 & 0.0017 & 0.0009 \\
n & 0.6123 & 0.1014 & 0.0142 & 0.0125 & 0.0147 & 0.0025 & 0.0010 \\
\hline
\end{tabular} \label{MSE3.5}
\end{table}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% latex table generated in R 3.4.3 by xtable 1.8-2 package
% Sun Jun 24 17:59:30 2018
\begin{table}[tbp]
\centering
\caption{$b_n$ $\times$ MSE of asymptotic variance estimators; $a_n = b_n^{2}$}
\begin{tabular}{rrrrrrrr}
\hline
$*$ & $b_n = 2^4$ & $b_n = 2^5$ & $b_n = 2^6$ & $b_n = 2^7$ & $b_n = 2^8$ & $b_n = 2^9$ & $b_n = 2^{10}$ \\
\hline
1 & 10.4229 & 21.6803 & 44.3604 & 87.4912 & 176.4313 & 354.4660 & 707.3375 \\
2 & 7.6130 & 18.6801 & 41.1353 & 83.6707 & 172.7787 & 350.4819 & 702.9933 \\
w & 3.3640 & 7.2915 & 17.8084 & 45.4855 & 101.8822 & 215.6920 & 477.9334 \\
m & 9.7003 & 12.7150 & 47.6132 & 82.8206 & 189.1759 & 322.4385 & 736.0519 \\
n & 21.3576 & 23.5368 & 20.4572 & 114.4274 & 352.5406 & 290.7497 & 678.8655 \\
\hline
\end{tabular} \label{bnMSE2.0}
\end{table}
% latex table generated in R 3.4.3 by xtable 1.8-2 package
% Sun Jun 24 17:58:54 2018
\begin{table}[tbp]
\centering
\caption{$b_n$ $\times$ MSE of asymptotic variance estimators; $a_n = b_n^{2.5}$}
\begin{tabular}{rrrrrrrr}
\hline
$*$ & $b_n = 2^4$ & $b_n = 2^5$ & $b_n = 2^6$ & $b_n = 2^7$ & $b_n = 2^8$ & $b_n = 2^9$ & $b_n = 2^{10}$ \\
\hline
1 & 10.6144 & 19.8530 & 41.7007 & 78.0909 & 136.5013 & 235.2049 & 380.0606 \\
2 & 7.7301 & 16.9243 & 38.3189 & 74.5957 & 133.4145 & 232.4373 & 377.8717 \\
w & 3.3050 & 6.7621 & 15.7652 & 37.3484 & 74.6952 & 134.0158 & 237.2865 \\
m & 9.3827 & 11.3368 & 41.9059 & 68.1650 & 140.6983 & 205.4797 & 385.9489 \\
n & 20.4606 & 20.2736 & 17.1543 & 91.9000 & 253.1752 & 168.8176 & 322.0486 \\
\hline
\end{tabular} \label{bnMSE2.5}
\end{table}
% latex table generated in R 3.4.3 by xtable 1.8-2 package
% Sun Jun 24 17:57:50 2018
\begin{table}[tbp]
\centering
\caption{$b_n$ $\times$ MSE of asymptotic variance estimators; $a_n = b_n^{3}$}
\begin{tabular}{rrrrrrrr}
\hline
$*$ & $b_n = 2^4$ & $b_n = 2^5$ & $b_n = 2^6$ & $b_n = 2^7$ & $b_n = 2^8$ & $b_n = 2^9$ & $b_n = 2^{10}$ \\
\hline
1 & 9.0823 & 13.9603 & 15.9025 & 13.4609 & 9.3177 & 5.7789 & 3.8676 \\
2 & 6.6952 & 11.7201 & 14.4583 & 12.7193 & 8.9962 & 5.6229 & 3.7991 \\
w & 2.9661 & 4.1263 & 4.7102 & 4.4094 & 2.7073 & 1.2673 & 1.3712 \\
m & 8.2390 & 7.1898 & 14.9288 & 10.2106 & 8.9112 & 3.5153 & 3.4378 \\
n & 17.9001 & 14.0135 & 5.2371 & 14.3203 & 22.6665 & 1.9175 & 1.7924 \\
\hline
\end{tabular} \label{bnMSE3.0}
\end{table}
% latex table generated in R 3.4.3 by xtable 1.8-2 package
% Sun Jun 24 17:56:42 2018
\begin{table}[tbp]
\centering
\caption{$b_n$ $\times$ MSE of asymptotic variance estimators; $a_n = b_n^{3.5}$}
\begin{tabular}{rrrrrrrr}
\hline
$*$ & $b_n = 2^4$ & $b_n = 2^5$ & $b_n = 2^6$ & $b_n = 2^7$ & $b_n = 2^8$ & $b_n = 2^9$ & $b_n = 2^{10}$ \\
\hline
1 & 5.2550 & 3.6621 & 2.7055 & 1.7321 & 1.4969 & 1.3790 & 1.4981 \\
2 & 3.9496 & 3.0341 & 2.4992 & 1.6330 & 1.4694 & 1.3785 & 1.4858 \\
w & 1.6055 & 0.9202 & 0.8457 & 1.0472 & 1.4993 & 2.1785 & 2.7900 \\
m & 4.4538 & 1.5082 & 2.0064 & 1.0981 & 1.0475 & 0.8577 & 0.8826 \\
n & 9.7961 & 3.2457 & 0.9092 & 1.5997 & 3.7559 & 1.2657 & 1.0496 \\
\hline
\end{tabular} \label{bnMSE3.5}
\end{table}
Table \ref{MSE2.0}, \ref{MSE2.5}, \ref{MSE3.0} and \ref{MSE3.5}
display the MSE of the asymptotic variance estimators
in the case that $a_n = b_n^r$ ($r = 2, 2.5, 3, 3.5$), respectively.
Table \ref{MSE2.0} and \ref{MSE2.5}
show that the MSE of the estimators do not converge to $0$.
Table \ref{MSE3.0} and \ref{MSE3.5} show that they converge to $0$.
These support Corollary \ref{avar} (a) and Corollary \ref{avar_k} (a).
Table \ref{bnMSE2.0}, \ref{bnMSE2.5}, \ref{bnMSE3.0} and \ref{bnMSE3.5}
display $b_n$ times the MSE of the asymptotic variance estimators
in the case that $a_n = b_n^r$ ($r = 2, 2.5, 3, 3.5$), respectively.
Table \ref{bnMSE2.0} and \ref{bnMSE2.5}
show that $b_n$ times the MSE of the estimators do not converge to $0$.
Table \ref{bnMSE3.0} and \ref{bnMSE3.5}
show that $b_n$ times the MSE of $\hat{\Xi}_{n,*}$ ($* = 1, 2, m, n$) converge to $0$,
and $b_n$ times the MSE of $\hat{\Xi}_{n,w}$ does not.
These support Corollary \ref{avar} (c) and Corollary \ref{avar_k} (c).
\section*{Acknowledgment}
The author would like to thank his supervisor Prof. Nakahiro Yoshida who provided valuable suggestions and comments.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\input{avar_estimator_ver6.1.bbl}
%\bibliographystyle{plain}
%
%\bibliography{bibtex20080401}
% BibTeX users please use one of
%\bibliographystyle{spbasic} % basic style, author-year citations
%\bibliographystyle{spmpsci} % mathematics and physical sciences
%\bibliographystyle{spphys} % APS-like style for physics
%\bibliography{bibtex-20150723-20150728-20151019} % name your BibTeX data base
%
%\bibliography{bibtex-since20170809}
\end{document}
\begin{thebibliography}{10}
\providecommand{\url}[1]{{#1}}
\providecommand{\urlprefix}{URL }
\expandafter\ifx\csname urlstyle\endcsname\relax
\providecommand{\doi}[1]{DOI~\discretionary{}{}{}#1}\else
\providecommand{\doi}{DOI~\discretionary{}{}{}\begingroup
\urlstyle{rm}\Url}\fi
\bibitem{2017arXiv170501446A}
{Abergel}, F., {Hur{\'e}}, C., {Pham}, H.: {Algorithmic trading in a
microstructural limit order book model}.
\newblock ArXiv e-prints (2017)
\bibitem{abergel2013mathematical}
Abergel, F., Jedidi, A.: A mathematical approach to order book modeling.
\newblock International Journal of Theoretical and Applied Finance
\textbf{16}(05), 1350,025 (2013)
\bibitem{abergel2015long}
Abergel, F., Jedidi, A.: Long-time behavior of a {H}awkes process--based limit
order book.
\newblock SIAM Journal on Financial Mathematics \textbf{6}(1), 1026--1043
(2015)
\bibitem{ait2010high}
A{\"\i}t-Sahalia, Y., Fan, J., Xiu, D.: High-frequency covariance estimates
with noisy and asynchronous financial data.
\newblock Journal of the American Statistical Association \textbf{105}(492),
1504--1517 (2010)
\bibitem{bacry2013modelling}
Bacry, E., Delattre, S., Hoffmann, M., Muzy, J.F.: Modelling microstructure
noise with mutually exciting point processes.
\newblock Quantitative Finance \textbf{13}(1), 65--77 (2013)
\bibitem{barndorff2011multivariate}
Barndorff-Nielsen, O.E., Hansen, P.R., Lunde, A., Shephard, N.: Multivariate
realised kernels: consistent positive semi-definite estimators of the
covariation of equity prices with noise and non-synchronous trading.
\newblock Journal of Econometrics \textbf{162}(2), 149--169 (2011)
\bibitem{10.2307/3598838}
Barndorff-Nielsen, O.E., Shephard, N.: Econometric analysis of realized
covariation: High frequency based covariance, regression, and correlation in
financial economics.
\newblock Econometrica \textbf{72}(3), 885--925 (2004).
\newblock \urlprefix\url{http://www.jstor.org/stable/3598838}
\bibitem{bibinger2011efficient}
Bibinger, M.: Efficient covariance estimation for asynchronous noisy
high-frequency data.
\newblock Scandinavian Journal of Statistics \textbf{38}(1), 23--45 (2011)
\bibitem{bibinger2012estimator}
Bibinger, M.: An estimator for the quadratic covariation of asynchronously
observed {I}t{\^o} processes with noise: asymptotic distribution theory.
\newblock Stochastic Processes and their Applications \textbf{122}(6),
2411--2453 (2012)
\bibitem{bowsher2007modelling}
Bowsher, C.G.: Modelling security market events in continuous time: Intensity
based, multivariate point process models.
\newblock Journal of Econometrics \textbf{141}(2), 876--912 (2007)
\bibitem{Chertok2016}
Chertok, A.V., Korolev, V.Y., Korchagin, A.Y.: Modeling high-frequency
non-homogeneous order flows by compound {C}ox processes*.
\newblock Journal of Mathematical Sciences \textbf{214}(1), 44--68 (2016).
\newblock \doi{10.1007/s10958-016-2757-6}.
\newblock \urlprefix\url{https://doi.org/10.1007/s10958-016-2757-6}
\bibitem{christensen2010pre}
Christensen, K., Kinnebrock, S., Podolskij, M.: Pre-averaging estimators of the
ex-post covariance matrix in noisy diffusion models with non-synchronous
data.
\newblock Journal of Econometrics \textbf{159}(1), 116--133 (2010)
\bibitem{christensen2013covariation}
Christensen, K., Podolskij, M., Vetter, M.: On covariation estimation for
multivariate continuous {I}t{\^o} semimartingales with noise in
non-synchronous observation schemes.
\newblock Journal of Multivariate Analysis \textbf{120}, 59--84 (2013)
\bibitem{CLINET20171800}
Clinet, S., Yoshida, N.: Statistical inference for ergodic point processes and
application to limit order book.
\newblock Stochastic Processes and their Applications \textbf{127}(6), 1800 --
1839 (2017).
\newblock \doi{http://dx.doi.org/10.1016/j.spa.2016.09.014}.
\newblock
\urlprefix\url{http://www.sciencedirect.com/science/article/pii/S0304414916301740}
\bibitem{cont2010stochastic}
Cont, R., Stoikov, S., Talreja, R.: A stochastic model for order book dynamics.
\newblock Operations research \textbf{58}(3), 549--563 (2010)
\bibitem{epps1979comovements}
Epps, T.W.: Comovements in stock prices in the very short run.
\newblock Journal of the American Statistical Association \textbf{74}(366a),
291--298 (1979)
\bibitem{GLOSTEN198571}
Glosten, L.R., Milgrom, P.R.: Bid, ask and transaction prices in a specialist
market with heterogeneously informed traders.
\newblock Journal of Financial Economics \textbf{14}(1), 71 -- 100 (1985).
\newblock \doi{http://dx.doi.org/10.1016/0304-405X(85)90044-3}.
\newblock
\urlprefix\url{http://www.sciencedirect.com/science/article/pii/0304405X85900443}
\bibitem{10.2307/2328052}
Gottlieb, G., Kalay, A.: Implications of the discreteness of observed stock
prices.
\newblock The Journal of Finance \textbf{40}(1), 135--153 (1985).
\newblock \urlprefix\url{http://www.jstor.org/stable/2328052}
\bibitem{griffin2011covariance}
Griffin, J.E., Oomen, R.C.: Covariance measurement in the presence of
non-synchronous trading and market microstructure noise.
\newblock Journal of Econometrics \textbf{160}(1), 58--68 (2011)
\bibitem{2011arXiv1106.5040G}
{Guilbaud}, F., {Pham}, H.: {Optimal High Frequency Trading with limit and
market orders}.
\newblock ArXiv e-prints (2011)
\bibitem{hayashi2008asymptotic}
Hayashi, T., Yoshida, N.: Asymptotic normality of a covariance estimator for
nonsynchronously observed diffusion processes.
\newblock Annals of the Institute of Statistical Mathematics \textbf{60}(2),
367--406 (2008)
\bibitem{hayashi2011nonsynchronous}
Hayashi, T., Yoshida, N.: Nonsynchronous covariation process and limit
theorems.
\newblock Stochastic Processes and their Applications \textbf{121}(10),
2416--2454 (2011)
\bibitem{hayashi2005covariance}
Hayashi, T., Yoshida, N., et~al.: On covariance estimation of non-synchronously
observed diffusion processes.
\newblock Bernoulli \textbf{11}(2), 359--379 (2005)
\bibitem{hewlett2006clustering}
Hewlett, P.: Clustering of order arrivals, price impact and trade path
optimisation.
\newblock In: Workshop on Financial Modeling with Jump processes, Ecole
Polytechnique, pp. 6--8 (2006)
\bibitem{jacod2009microstructure}
Jacod, J., Li, Y., Mykland, P.A., Podolskij, M., Vetter, M.: Microstructure
noise in the continuous case: the pre-averaging approach.
\newblock Stochastic processes and their applications \textbf{119}(7),
2249--2276 (2009)
\bibitem{Kimura2016}
Kimura, A., Yoshida, N.: Estimation of Correlation Between Latent Processes,
pp. 131--146.
\newblock Springer International Publishing, Cham (2016).
\newblock \doi{10.1007/978-3-319-45875-5\_6}.
\newblock \urlprefix\url{https://doi.org/10.1007/978-3-319-45875-5\_6}
\bibitem{koike2013estimation}
Koike, Y.: Estimation of integrated covariances in the simultaneous presence of
nonsynchronicity, microstructure noise and jumps.
\newblock Econometric Theory pp. 1--79 (2013)
\bibitem{koike2014estimator}
Koike, Y.: An estimator for the cumulative co-volatility of asynchronously
observed semimartingales with jumps.
\newblock Scandinavian Journal of Statistics \textbf{41}(2), 460--481 (2014)
\bibitem{koike2014limit}
Koike, Y.: Limit theorems for the pre-averaged {H}ayashi--{Y}oshida estimator
with random sampling.
\newblock Stochastic Processes and their Applications \textbf{124}(8),
2699--2753 (2014)
\bibitem{2014arXiv1410.1900K}
{Korolev}, V.Y., {Chertok}, A.V., {Korchagin}, A.Y., {Zeifman}, A.I.: {Modeling
high-frequency order flow imbalance by functional limit theorems for
two-sided risk processes}.
\newblock ArXiv e-prints (2014)
\bibitem{large2007measuring}
Large, J.: Measuring the resiliency of an electronic limit order book.
\newblock Journal of Financial Markets \textbf{10}(1), 1--25 (2007)
\bibitem{malliavin2002fourier}
Malliavin, P., Mancino, M.E.: Fourier series method for measurement of
multivariate volatilities.
\newblock Finance and Stochastics \textbf{6}(1), 49--61 (2002)
\bibitem{Malliavin2007}
Malliavin, P., Mancino, M.E., Recchioni, M.C.: A non-parametric calibration of
the {HJM} geometry: an application of {I}t{\^o} calculus to financial
statistics.
\newblock Japanese Journal of Mathematics \textbf{2}(1), 55--77 (2007).
\newblock \doi{10.1007/s11537-007-0666-7}.
\newblock \urlprefix\url{https://doi.org/10.1007/s11537-007-0666-7}
\bibitem{malliavin2009fourier}
Malliavin, P., Mancino, M.E., et~al.: A {F}ourier transform method for
nonparametric estimation of multivariate volatility.
\newblock The Annals of Statistics \textbf{37}(4), 1983--2010 (2009)
\bibitem{mancino2011estimating}
Mancino, M.E., Sanfelici, S.: Estimating covariance via {F}ourier method in the
presence of asynchronous trading and microstructure noise.
\newblock Journal of financial econometrics \textbf{9}(2), 367--408 (2011)
\bibitem{2010arXiv1003.3796M}
{Muni Toke}, I.: {``Market making'' behaviour in an order book model and its
impact on the bid-ask spread}.
\newblock ArXiv e-prints (2010)
\bibitem{2015arXiv150203871M}
{Muni Toke}, I.: {Stationary distribution of the volume at the best quote in a
Poisson order book model}.
\newblock ArXiv e-prints (2015)
\bibitem{muni2011modelling}
Muni~Toke, I., Pomponio, F.: Modelling trades-through in a limited order book
using {H}awkes processes.
\newblock Economics discussion paper (2011-32) (2011)
\bibitem{2016arXiv160203944M}
{Muni Toke}, I., {Yoshida}, N.: {Modelling intensities of order flows in a
limit order book}.
\newblock ArXiv e-prints (2016)
\bibitem{mykland2012gaussian}
Mykland, P.A.: A {G}aussian calculus for inference from high frequency data.
\newblock Annals of finance \textbf{8}(2-3), 235--258 (2012)
\bibitem{ogihara2015quasi}
Ogihara, T., Yoshida, N.: Quasi likelihood analysis for point process
regression models.
\newblock preprint (2015)
\bibitem{park2012estimating}
Park, S., Linton, O.B., et~al.: Estimating the quadratic covariation matrix for
an asynchronously observed continuous time signal masked by additive noise
(2012)
\bibitem{podolskij2007estimation}
Podolskij, M., Vetter, M., Sommer, M.: Estimation of volatility functionals in
the simultaneous presence of microstructure noise and jumps.
\newblock Available at SSRN 950344 (2007)
\bibitem{10.2307/2327617}
Roll, R.: A simple implicit measure of the effective bid-ask spread in an
efficient market.
\newblock The Journal of Finance \textbf{39}(4), 1127--1139 (1984).
\newblock \urlprefix\url{http://www.jstor.org/stable/2327617}
\bibitem{smith2003statistical}
Smith, E., Farmer, J.D., Gillemot, L.s., Krishnamurthy, S., et~al.: Statistical
theory of the continuous double auction.
\newblock Quantitative finance \textbf{3}(6), 481--514 (2003)
\bibitem{doi:10.1080/14697688.2014.963654}
Toke, I.M.: The order book as a queueing system: average depth and influence of
the size of limit orders.
\newblock Quantitative Finance \textbf{15}(5), 795--808 (2015).
\newblock \doi{10.1080/14697688.2014.963654}.
\newblock \urlprefix\url{http://dx.doi.org/10.1080/14697688.2014.963654}
\bibitem{voev2007integrated}
Voev, V., Lunde, A.: Integrated covariance estimation using high-frequency data
in the presence of noise.
\newblock Journal of Financial Econometrics \textbf{5}(1), 68--104 (2007)
\bibitem{zhang2012tale}
Zhang, L., Mykland, P.A., A{\"\i}t-Sahalia, Y.: A tale of two time scales.
\newblock Journal of the American Statistical Association (2012)
\bibitem{zhang2006efficient}
Zhang, L., et~al.: Efficient estimation of stochastic volatility using noisy
observations: A multi-scale approach.
\newblock Bernoulli \textbf{12}(6), 1019--1043 (2006)
\bibitem{zhou1996high}
Zhou, B.: High-frequency data and volatility in foreign-exchange rates.
\newblock Journal of Business \& Economic Statistics \textbf{14}(1), 45--52
(1996)
\end{thebibliography}