Hi, In order to produce an autoregression where only certain lags are allowed, specified in advance (e.g. c(1,2,5) ), I have found it necessary to look beyond the standard [ar] function, thankfully discovering the [FitAR] package, wherein the [FitARp] function provided exactly that capability. However for my problem at hand, [FitARp] is vastly slower than [ar] - taking hours rather than minutes. A lesser problem is that it recognises only vectors and matrices, not [zoo] objects etc. So I wonder: 1) Is there a further alternative with that capability? 2) Is there any way to derive a plain [ar] model from a [FitARp] model? I imagine that should be possibe in principle, it should be just a case of some coefficients being zero. Any help/advice would be appreciated. David Esp, UK. [[alternative HTML version deleted]]