LAMN property for hidden processes: the case of integrated diffusions

DOI10.1214/07-AIHP111MaRDI QIDQ731453zbMATH OpenOpenAlexFDO

Publication date 7 October 2009

Published in Annales de l'Institut Henri Poincaré. Probabilités et Statistiques (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/0707.0257, https://eudml.org/doc/77957

conditional information integrated diffusion processes likelihood ratio local asymptotic mixed normality

Stochastic calculus of variations and the Malliavin calculus (60H07) Diffusion processes (60J60) Asymptotic properties of parametric estimators (62F12) Non-Markovian processes: estimation (62M09)

Abstract: In this paper we prove the Local Asymptotic Mixed Normality (LAMN) property for the statistical model given by the observation of local means of a diffusion process

X

. Our data are given by

i n t_{0}^{1} X_{f r a c s + i n} d d m u (s)

for

i = 0, ..., n - 1

and the unknown parameter appears in the diffusion coefficient of the process

X

only. Although the data are nor Markovian neither Gaussian we can write down, with help of Malliavin calculus, an explicit expression for the log-likelihood of the model, and then study the asymptotic expansion. We actually find that the asymptotic information of this model is the same one as for a usual discrete sampling of

X

.

Review text

A diffusion process \(X_t\) is considered satisfying the (scalar) equation \[ dX_t=a(X_t,\vartheta)dB_t+b(X_t)dt, \] \(B\) being a Brownian motion, where \(\vartheta\in \mathbb R\) is an unknown parameter. The observations are integrated, i.e., \[ \bar X_j=\int_0^1 X_{(s+j)/n}d\mu(s)\quad \text{for } j=0,\dots,n-1, \] where \(\mu\) is some fixed known measure on \([0,1]\). The authors demonstrate that the likelihood ratio of the observations satisfys the local asymptotic mixed normality (LAMN) property with the conditional information \(I_\vartheta=2\int_0^1 (a'_\vartheta/a)^2(X_s,\vartheta)ds\), which is the same for the conditional information of non-integrated observations \(X_{j/n}\), \(j=0,1,\dots,n-1\). So the asymptotically optimal estimates for \(\vartheta\) have the same convergence rates for integrated and non-integrated observations. The proof is based on an explicit expression for the log-likelihood derived by a Malliavin calculus technique.

Recommendations

Cites work

Cited in

(16)

This page was built for publication: LAMN property for hidden processes: the case of integrated diffusions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q731453)