I believe the logical reason is this:
When you make a track, you have no idea about how loud it will be played back by the listener. But you need some kind of relative measure, so they came up with this idea to take a maximum amplitude for the signal distribution (+- 1 volt I think, 16/24/32 bits digitally), name it 0db and base every measuremet out of it. (Thus logically you go down.) That maximum is in the end controlled by a listener's volume knob. (When he sets the volume to 100db, your -5db peaking track will be 95db loud in reality.)