My client has a portfolio (all sort of derivatives) for which he wants to get a VAR figure .. we are approaching it by segregating them on their type .. for swaptions for instance we had a pricer which took a daily change since 2005 for forward swap rate, OIS curve and vol . and using this we created a price series for each swaption ..
however, for futures, I am still not clear which way would be better .. also as u said that if we stitch them using spread the resultant new curve is a lot different than the original one .. which hampers the daily return calculation .. as far as i can remember BBG gives u a couple of ways for this roll over adjustment .. difference(the one u suggested), ratio , average etc .. i found the avg one better suited since it takes a proportion of both active and 2nd generic contact and the shape is pretty much undistorted from the original ..
and comments on this . also would be useful if u can refer me any research paper ..