As the others say, the submission format of the paper should be whatever you think will make it easiest for the referees. I like to number every equation, for example, so the referee can refer to them, even though a final version for the reader shouldn't number every single one. I also like plentiful steps of algebra, even though that isn't good for the final reader.
There are some exceptions. AER, if I remember rightly and am up to date, will send your paper back to you if it isn't double-spaced, which I wouldnt bother with unless a journal insists. You won't get a rejection for that, just a bounceback and a month's delay.
You might think that taking a lot of trouble to format your paper to the particular journal's style will signal that you really think the paper is good, and as a credible signal, something you wouldn't do if you were pretty sure they'd reject, it would help your chances of acceptance. Wrong. The signal is too weak, and the unobservable qualities of the paper too unimportant (or, put differently, the referees will not think that your ability to know whether your paper is good is accurate enough to be worth knowing). Also, there are plenty of people who think their paper is good but have their own preferred formatting and aren't going to reformat to what they think is inferior until they have to for the final version (e.g., I've always used the "Rasmusen (1989)" format for citations instead of the stupid "[10]" system some science journals use.) Finally, the referees are almost 100% certain not to even remember what style the journal they are late refereeing for uses.
On the other hand, good writing style is important in itself and improves your chances of acceptance by making life easier for the referee and preventing him from falling into a sour and bitter mood and having to suppress his desire for vengeance against you for writing so confusingly.