look at cyclic variability models. You transform the dates into something like degrees on the compass and then do the regression. It allows you to look at various sinusoidal type patterns. It helps to have a postulated model ( days of the week, say). I'm rusty on it, but there was a good bit of literature on it.