Ulf Moehring schrieb: Well I was waiting for sb else to answer this... Arima models are already pretty complicated internally and inherently nonlinear. Therefore I'd say the usual statistics would be only valid asymptotically, which weakens the case for Adjusted R^2 instead of regular R^2. Also, for given samples it may not be obvious which formula is the most adequate. For example gretl usually will report that the estimated residuals/innovations have a non-zero mean, because the usual geometry of linear models does not apply and so residuals will not have exact mean zero. So the usual R^2 formula would not be 100% appropriate in my view. Having said that, gretl reports the means and variances of the residuals and of the dependent variable, so you definitely could quite easily calculate something once you've decided which formula to use. And of course, it is possible that reporting some additional statistics for Arima models would be a good thing for future gretl versions (properly documented, of course), but right now I don't know if that's true. cheers, sven