Agreeing with all of the advice above, none of it took explicitely explicitly into account the time series aspect of the data. If you are using models (like arima) where the time dependence is axplicit explicit in the model, it is not possible to use standard cross validadtionvalidation.
What I have done in such caeses cases is choose a model, estimate it using the 5 firts yesars first years of data, then predict the five next years, getting a discrepancy measure, and continuing like this.
This makes much practical value.
