i'm using ident toolbox in matlab 2007a. i get moderate results using an ARX model in simulation mode but if i switch to prediction mode, the results swing to almost 99%. however, even when a model is 99% accurate, the residual analysis results were really bad. both tests (white and indep) did not turn out well.
how can a good model still yield bad residual tests results. is there any meaningful relationship between the two information sources? ie. model output and residual analysis