The method includes (or implies) how to calculate the running variance. (See my response to Andrew.)
The value you want is the running variance at the center of each year. (I haven't thought about how to extend it to cases where the length of the year isn't constant.)
Hope this helps,
Ryo