I got straight A's in economics in college (along with A's in science/math/computers, but F's in most all other things - if it's not interesting to me, I can't make myself do it it seems).
I never practiced it, outside how I use it to guide my trading - and that's not quantitative at all, I just use it to guide my gut.
My takeaway was this - yes, some fairly simple rules can tell you quite a lot, and be very predictive - BUT! (and it's a huge hairy ugly butt) - you have to have the right numbers to plug in and the details really matter, like with any equation-set that can go divergent through iteration (fractals).
Thus, it's just like predicting the weather, which only takes 6 or so pretty simple equations. BUT! For the weather, this means all you need is 6-7 values measured for every small packet of air in a rather large area - you need both resolution and scale, and it's very dangerous to interpolate any holes in the data (which might contain a microburst - or a tornado). Then you can just grind through this simple stuff (but a lot of it!) and get to the weather on the next roundy round - any errors in original data or the calculation of course can do worse than add up - that's the basic thing we mean when we say "divergent". Also, errors in assumptions at the sample border (no one has yet managed to get a full snapshot of the entire atmosphere) can come screaming into the middle of your set real quick.
Economists don't have these numbers. It's pretty hopeless for them to ever get them, and they all matter, from what the crack dealer on the corner gets on up - they ALL are important, and the black market (far larger than estimated), barter, cooked books, and so on mean they'll never have them.
But I guess once you pay all that money for schooling, you have to do something to eat and pay back the student loans anyway. There are an awful large number of other PhD's in the same boat it seems these days.
A lot of it comes down to the fact that there are as yet, no closed form (feedforward) solutions to quite a number of apparently simple problems - it's not just economics.
The three body problem in gravitation is one that comes to mind in physics. There is no math you can plug in say, the masses, speeds, locations of all the stuff in the solar system and ask "where is Jupiter in 100 years". You can only calculate where it's going to be in a millisecond - everything moved, so all the numbers for the next roundy round have changed, repeat till you get to 100 years. You can see how utterly error-prone this is going to be with any finite time slice above zero, or any error in initial conditions.
Folding@home is facing the same set of issues. It's all over science of all kinds.
You can tell what's going to happen in some simple case next, but that's it.
The general invert-ability problem is another. You can give me a chemical lattice structure, and if need be I can go all the way down to the wavefunctions of the pieces and generate all the properties this thing will have (at least, the simple physical ones, not perhaps how it would interact with other things as a drug).
But! There's no way to say - gee, I need something that's this strong, with that melting point, strain rate - and then use that math to design the compound.
The more you know, the more you know you don't know...still true.