Why does the boundary value analysis provide good test cases?

Because errors are frequently made during programming of the different cases near the 'edges' of the range of values.