Maybe the previous weeks have an effect on the next, for example a low price point brings more new users that will bias the result of the next week and so forth. It's hard to make a controlled experiment... but at least it is possible to do multiple experiments over time shuffling the order in which the prices are tried.