We present a technique for building deterministic models of the nonlinear d
ynamics underlying observed time series. It is formulated from maximum entr
opy principle within the framework of information theory. Two numerical exa
mples of chaotic time series illustrate the method. Furthermore, the theory
on which the method is based, provides an entropy-like quantity that chara
cterizes the suitability of the model, It is defined over the space of para
meters of the model. We illustrate, with two applications, how this entropy
can be a useful tool for exploring the phase space, and establishing a cri
terion for choosing a convenient parametrized functional form. (C) 2000 Els
evier Science B.V. All rights reserved.