C++ :: Type Promotion Rules For Fixed Point Class

I've come across an interesting question involving a fixed-point arithmetic class.

Suppose I've got a template that implements a fixed-point quantity. It has two characteristics: width and number of fraction bits:

Code:
template <typename RAWTYPE, unsigned int FRACBITS>
class Fixed
{
....
};

So if I want a 32-bit value with 12 bits of fraction I can declare:

Code:
Fixed<int32_t, 12> x;

Suppose I've implemented a freestanding operator+() for fixed values, and I want to be able to add different types together:

Code:
template <typename RAWTYPE1, unsigned int FRACBITS1, typename RAWTYPE2, unsigned int FRACBITS2>
???? operator+(Fixed<RAWTYPE1, FRACBITS1> lhs, Fixed<RAWTYPE2, FRACBITS2> rhs)
{
???
}

What is the result type? Obviously, it's up to me to decide this. As reference, consider the type promotion rules for native types:

Code:
short a;
int b;
int result = a + b; In this case, the short value is promoted to the int value, and the addition happens on int.

It would seem a similar rule (go to the wider type) would be appropriate for fixed point. But there is another dimension to the problem, which is the number of fraction bits. Should you go to the wider type? Or the most precise type? Should you endeavor to minimize the number of bits which are discarded? What's the most intuitive rule?