High-frequency analysis of parabolic stochastic PDEs (Q2196213)

This paper deals with the estimation of the noise process in an observed process defined through a parabolic stochastic partial differential equation. The author proposes a realistic mathematical model regarding the various applications that this process has in various disciplines. Let's see a brief introduction to paper math. Be $(\Omega,F,P)$ a probability space and be $(F_{t})_{t\in\mathbb R}$ a filtration on $(\Omega,F,P)$, that is: $F_{s}\subset F_{t}\subset F$ for $s\leq t$, and $F_{s}=\cap_{s\leq t}F_{t}$ for $s\in\mathbb R$. For each $(t,x)\in\mathbb R\times\mathbb R^{d}$ $(d\geq1)$ be $Y(t,x):\Omega\to\mathbb R$, $\sigma(t,x):\Omega\to\mathbb R$ and $W(t,x):\Omega\to\mathbb R$ random variables such that the below parabolic stochastic PDE is satisfied \[ \partial_{t}Y(t,x)=(\kappa/2)\Delta Y(t,x)-\lambda Y(t,x)+\sigma(t,x)W(t,x),\tag{1} \] where $W=(W(t,x))_{(t,x)\in\mathbb R\times\mathbb R^{d}}$ is a Gaussian noise and $\Delta$ is the Laplacian operator on the spatial coordinates of the process $Y=(Y(t,x))_{(t,x)\in\mathbb R\times\mathbb R^{d}}$. In practice, $Y$ is the observed process, $\kappa>0$, $\lambda>0$ and the process $\sigma$ are estimated or some of them are considered known. This paper proposes a consistent estimator and asymptotic confidence limits for the process $\sigma$ called stochastic volatility process, not only in the context of finance. Here it is said that, if \[ \sup_{(t,x)\in\mathbb R\times\mathbb R^{d}} E\left[\sigma^2(t,x)\right] < \infty, \] then (1) has a solution given by \[ Y(t,x)=\int_{\mathbb R^{d}}\int_{-\infty}^{t}G(t-s,x-y)\sigma(s,y)\cdot W(ds,dy), \] where \[ G(t,x)=(2\pi\kappa t)^{-d/2} \exp(-\Vert x\Vert^2/(2\kappa t)-\lambda t) \cdot \1_{(0,+\infty)}(t). \] It is assumed that the process $Y$ is observed in a finite number of $\mathbb R^{d}$ and in a much larger number of times $t=\Delta_{n},2\Delta_{n},\dots,[(T/(\Delta_{n}))]\Delta_{n}$ within the interval $[0,T]$, $T<\infty$ ($[a]$ is the integer part of $a\in\mathbb R$). The proposed scheme of observations, of low spatial resolution and high frequency in time, is a realistic approach in many applications. The authors study properties of the solution to Equation~(1) for $\Delta_{n}\to 0$. This requires prior knowledge of two concepts: ``uniform convergence in probability over compacts'', and ``stable functional convergence in law with respect to uniform topology''. These concepts and basic results related to them can be studied in the following two books that are in the bibliography of this paper: [\textit{Y. Aït-Sahalia} and \textit{J. Jacod}, High-frequency financial econometrics. Princeton, NJ: Princeton University Press (2014; Zbl 1298.91018)] and [\textit{J. Jacod} and \textit{P. Protter}, Discretization of processes. Berlin: Springer (2012; Zbl 1259.60004)]. Thorough knowledge of these two texts is essential for reading this paper. In several remarks, the author comments on the need for certain suppositions and relaxed approaches in order to extend the results stated in the paper to other models. In this paper, the author does not give the proof of any of the several results that he enunciates by means of lemmas and theorems. He only gives various comments on them and he says that the proofs are in another ``Supplement'' paper that is not provided by the author free of charge. No examples of application of the results to real or simulated situations are given either. It is hard to understand this work without good knowledge, typical of a mathematician, about stochastic differential equations.