We are trying to produce partial vega using our BGM model. Even though the total/parallel vega is correct, the partial vega numbers (e.g. vega for each swaption vol cube bucket) do not make sense due to the vol surface interpolation and re-calibration when the buckets are bumped. I am looking some literature on this topic... It looks like that mathematically this problem is more general: e.g. similar issues also arise when calculating delta on a portfolio - e.g. how to convert delta risk from one set of tenors to another (this depends on the tenors specification, interpolation method used, etc)Thanks, Michael

There is a paper by Pietersz for Bermudan vegahttp://papers.ssrn.com/sol3/papers.cfm?abstract_id=383580I talk about it a bit herehttp://papers.ssrn.com/sol3/papers.cfm?abstract_id=427084-V

Thanks!Michael