It allows to build a history. When I got five different datasets about you from five different sources which list three different addresses, I know that you seem to be someone who moves a lot, which can have various reasons. When I can pinpoint that a while ago your last name changed and shortly afterwards your address changed as well, it tells me you either married or got divorced. So I will spam you either with household items or dating sites.
It gives me more material to link datasets together. When I have one dataset from you with an outdated telephone number and another old dataset which is anonymous except for the same outdated telephone number, I can link them to each other.
It also allows identity theft in cases where personal information is used to identify you. For example, you might have some old email account you don't use anymore. When I would hack that account somehow, you wouldn't even notice. But maybe there are still some websites where you have accounts bound to that old email account. I could now use the "forgot password" function of these websites to take over your accounts.
However I still agree that less data is better than more and that one should generally avoid to have more data about them in some databases than strictly necessary.