A. Elmoataz et al. (Eds.): ICISP 2012, LNCS 7340, pp. 326–332, 2012.
© Springer-Verlag Berlin Heidelberg 2012
Maximum Likelihood Estimation, Interpolation
and Prediction for Fractional Brownian Motion
Rachid Harba
1
, Hassan Douzi
2
, and Mohamed El Hajji
2
1
Laboratoire PRISME, Polytech’Orléans, Université d'Orléans 45067 Orléans, France
Rachid.Harba@univ-orleans.fr
2
Laboratoire IRF-SIC, Faculté des sciences, Université Ibn Zohr, BP8106 Agadir, Maroc
{douzi_h,hajjimohmed}@yahoo.fr
Abstract. The maximum likelihood (ML) estimation approach for fractional
Brownian motion (fBm) is explored in this communication. First, a ML based
estimation of the H parameter is implemented on the signal itself. This approach
on the signal itself can easily be applied on non-uniformly sampled data or
directly useful in the case of incomplete data. Secondly, the method is extended
to provide a ML prediction and a ML interpolation for fBm which could be of
interest in many domains. Results also help to explain errors in other
interpolating methods such as the midpoint displacement algorithm used to
synthesize fBm data.
1 Introduction
Fractional Brownian motion (fBm) of H parameter in the range ]0 ; 1[ is defined as an
extension of Brownian motion [1]. One of the main issues when dealing with such
data is to estimate the H parameter [2-3]. Among the numerous methods to achieve
such a goal, the maximum likelihood (ML) approach proposed by Lundahl et al. [4] is
often used due to its asymptotical efficiency [5]. It is also efficient in noisy
environments [6]. But the ML based estimation of the H parameter is performed on
the fBm increments which may be a limiting factor in some cases.
Here, we propose an ML estimate of the H parameter processed on the fBm itself.
This allows direct extension of the method to include cases where there may be
irregular sampling or incomplete data. Moreover, an ML based prediction and
interpolation technique for a fBm signal easily result.
This communication is organized as follow. In the next section, fBm is defined and
its main properties are derived. Then, the ML based estimation of the H parameter is
achieved and is tested on exact fBm data. Finally, ML interpolation and prediction are
presented and a real data example illustrates the methods.
2 FBm Properties
Continuous fBm of H parameter in ]0 ; 1[, denoted B
H
(t), is defined as an extension of
Brownian motion B(t) [1]: