Optimal linear filtering and smoothing for a discrete-time stable linear model (Q1333192)

This paper studies the optimal linear filtering, smoothing, and prediction problem for a discrete-time stable linear model with constant coefficients. In Section 2, some technical results, conditional expectation and metric projection for multivariate symmetric stable distributions are established. In Section 3, the author considers the symmetric stable model and generalizes the classical Kalman-Bucy discrete-time Gaussian model; Theorem 3.1 gives a recursive formula for the optimal linear filtered estimate and for the estimation error. In Section 4, we can find the main result of this paper: Theorem 4.1 gives an explicit formula for the optimal linear smoothed estimate of the signal \(X_ k\) given not only the whole past but also a part of the future observations \((Y_ 1, \dots, Y_{k + m})\). Finally Proposition 4.2 gives a recursive formula for the optimal linear predicted estimate, and its conditional expectation, of the signal \(X_{k + m}\) based only on the part of the past observations \((Y_ 1, \dots, Y_ k)\). A nice paper, and I think that its extension to continuous-time model will be very interesting.