That's a good test but hardly sufficient. I had to do stuff like that with business models in a course for my MBA degree. I'd find econometric variables that correlated with what I was trying to predict and then write a paper showing how it worked well for past instances. Now, if I really KNEW that it would work going forward, I coulda made some bucks in the market.

Bottom line, it's a necessary but not sufficient condition that the model predict past behaviour. What would indicate a really good model would be if it output some factor about a past climate that later research revealed to be true without us having known it beforehand. That kinda model would have some credibility.