Hello -
Negative time courses could mean any number of things depending on your paradigm, ROIs, method of extracting the percentage signal etc.
It could also mean that the voxels in the ROI you are looking at operates at a tonically high level during the baseline condition and when an experimental condition forces it to switch off. Given that you have periodicity and not negative 'noise' this may be the case. Marcus Raichles stuff is a must read here - most issues of PNAS during the last 5 years have one of his articles on baseline responses. Jason Mitchell from Dartmouth also published a nice PNAS paper the other month on this. If you cant find it drop me a line and i will email to you.
carl