Computationally efficient stochastic realization for internal multiscale autoregressive models (Q5935450)

The paper describes computationally efficient \(O(N^{2})\) algorithms for generating multiscale autoregressive (MAR) processes. More explicitly, one considers the problem of selecting the parameters for the dynamics of a MAR process to model the second-order statistics of any given fine-scale one- or multi-dimensional random process. The theory focuses on the estimation-theoretic concept of predictive efficiency and refers to the scale-recursive structure of the so-called internal MAR process. For the help of the reader, the paper is self-contained, and many prerequisites (MAR processes, internal model, Markovianity, predictive theory) are reviewed. And then one describes two algorithms: one with complexity quadratic in problem size and another one with complexity linear in problem size.