Recurrence relations can be used for a lot of things, not only complexity estimation.
But as for complexity estimation you are surely misusing them when you get a recurrence relation such as T(n) = T(n-1) - T(n-2).
Of course you can write a program such as:
function fib(int n)
{
if (n == 0) return 1;
if (n == 1) return 1;
return fib(n-1) - fib(n-2);
}
but note that you will still have to add up calls from fib(n-1) to fib(n-2), because you call them, so your recurrence relation for T(n) = T(n-1) - T(n-2) doesn't make sense for programs.
As for the example given by Eric J., he gave a formula that calculates something, however the complexity of the program I gave above is still O(2^n), because you have to add operation counts, as svs said in the comments.
fib(n)
/\
fib(n-1) fib(n-2)
/\ /\
fib(n-2) fib(n-3) fib(n-4)
...............................
The above is a function calling graph for both cases regardless of whether there's a minus in between or not.
PS: I didn't mean to have 2 edges going into fib(n-3), i wanted to write fib(n-3) separately, but I guess it looks better this way.
The program above does not calculate Fibonacci sequence, I only modified it by putting a minus there.