Making the creator's beliefs (including hidden assumptions) completely clear is valuable all by itself. But in addition to revealing the creator's beliefs it also reveals arithmetic errors, which are quite common in poorly specified verbal models.

For an example of the latter, observe this HN thread: https://news.ycombinator.com/item?id=7042469

You may find this interesting:

http://www2.econ.iastate.edu/tesfatsi/WhyEconomistsShunSimul...

That's all any model used to construct a predictive simulation can do.