The only reasonable thing I could weigh in as being an advantage of writing your own is the knowledge gained in the experience. Reading a framework's source and writing your own are two entirely different experiences. Even in that situation though, it still comes down to how much you are willing to ignore pre-existing constructs and standards. I say reasonable, because there is the off chance that you might be a genius, or lucky, and come up with something amazing through your self imposed limits on influence. Your inventive potential stands to gain much from ignoring the status quo, should you even take it that far.
As for why you shouldn't just make something that makes sense to you, well: what if you spent some time making some pre-existing framework make sense to you? Could you perhaps save time in the long run, compared to your doing your own thing? Or if time isn't your concern, then insert whatever it is you are trying to get out of this into that question and think long and hard.
Basically, if writing your own framework is something you're seriously considering, you stand to become intimately familiar w/the material which is good if you are specializing in the field. If you feel your ingenuity is resilient enough to not miss out on any great ideas by immersing yourself in extant APIs. Alternatively, if you are unsure of your actual familiarity with a subject in the face of everything that is out there, taking Cake for a spin might do you a world of good and give you more ideas than you'd come up w/otherwise. Perhaps you may find that you can even improve upon one that already suits your tastes. :)
