Just the other day I thought the same thing, but then I thought, "as a sanity check, why don't I look at the issue in frequency space?" So, I transformed the data to its inverse space and verified my conclusions. I have to admit the functional dependence of the phase shift was surprising.