Numerical Evaluation Of Distributions In Non‐Linear Autoregression

Abstract. We use the Chapman‐Kolmogorov formula as a recursive relation for computing the m‐step‐ahead conditional density of a non‐linear autoregressive model. We approximate the stationary marginal probability density function of the model by the m‐step‐ahead conditional density for sufficiently large m. An advantage of our method is its simple implementation; only one NAG subroutine is needed. We have also studied the advantage of incorporating the matrix‐squaring procedure.