Autoregression methods in Macroeconometrics

So, there's a lot of discussion about what economists should be doing. Are we merely to report? Explain? Forecast? Or (heaven forbid) prescribe? Anyways, one that I've always taken for granted as a necessary part of economic is the "explain" part. That should be the point of any area of study - understand and explain the 'why' of what's going on - because everything else requires and understanding of the basics. Medicine, for example, would be worthless if doctors didn't understand the whys of diseases and maladies. Now, some people may disagree, saying that we can often fix things without truly understanding why it works. To those people, I have to say that you're just guessing. If you prescribe something without understanding why it works, you have no idea what you're doing.

Anyways, back to economics. Much of macroeconomics is about trying to understand why economies behave the way they do. Macroeconomists study why (or if) countries converge in output, how quickly developing countries develop and why, what effect money has on output, and all sorts of other (interesting) things. So, we take data (and lots of it) and try to make functions for output, growth, and all sorts of fun things. But then came along a group of economists in the 1970s who said, "All you (stupid) economists from 1950 until today who were trying to explain why output changes? Well, you're prediction powers are crappy at best. But look! If we use no explanatory variables, we can do such a better job than you, you Keynesians!"

And along came the Box-Jenkins ARIMA method, which only uses lagged measurements of the dependent variable to explain, yep, that dependent variable.

So my question, is this: so what? Yay, you can much more accurately explain what's going on this year by looking at last year, but why? What could possibly be the point of a regression like that, besides an exercise in saying, "Look how close I can get while explaining absolutely nothing!"

Or am I the only one that thinks explanations (and therefore explanatory variables) are important?

It's like saying in medicine, "Let's explain how healthy this patient is based on how healthy they've been in the past!" Or in psychology, "We can guess how this person will behave based on how they've behaved in the past!"