Choose data from the most recent two years for out of sample testing

Question

I have a dataset that contains monthly data for the past four decades. I understand that if I use 5-fold cross validation, I need to randomly pick training set and testing set. Can I use data from the most recent two years as the testing set? If so, how should I revise the procedure? Thanks!

Answer 1

Yes you can, since standard CV doesn't work for time series data. See: https://www.otexts.org/fpp2/accuracy.html