That would depend on what the starting value of X is, which is no different than seeding the rand function. Additionally, any PRNG can be manipulated if the algorithm of the PRNG is known. You also can check the distribution of your function with enough trials, and it can be seen that the function is strictly decreasing along the negative half of a triangular distribution. This means that no matter what X starts as, it will eventually give an overflow error as it gets too large; so, it isn't exactly a PRNG since given any input, you know that the next input will be more negative and that no two values will occur twice
While this may work for some applications, I personally don't feel that it classifies as a PRNG, as it doesn't create a probability distribution when the number of iterations goes to infinity