Selecting nonlinear stochastic process rate models using information criteria (Q2492254)

In spite of the widespread use of stochastic process models, techniques which connect them to observational data are somewhat limited, so that methods are needed to estimate parameter values from data and to perform model selection, that is, to select from some defined class the model which is best supported by the data. The authors' aim is to demonstrate how unknown process rates within a stochastic modelling framework based on Markov processes can be approximated from time series data using polynomial functions. The problem of model selection is addressed by adapting basis function selection methods and minimum description length information criteria that have previously been developed for nonlinear autoregressive models of time series under Gaussian noise assumptions. The authors focus on the case of complete data. \noindent It is nevertheless anticipated that much of the apparatus developped in this case will be applicable when only incomplete data are available. The results obtained are illustrated using two idealized examples from population biology.