January 4th, 2011, 12:37 pm
ok, thinking aloud, I think your major exposure is to smile term structure + underlying vol correlations. So any model that can handle this should be fine. I think if you can have a decent price back for variance swaps (rather vol swap) then the other thing, i.e. vol/underlying correl becomes more important. So I would be happy with a stochvol term structure model. You say you get significantly different values using LV and Bergomi, do they had same implied autocorrels and vol/underlying correls?Also this would be perhaps doable only for the shorter end, else the heavy computation might be problematic for risk runs, so I think one way is to split the price in to, say a 3m horizon with a propoer MC, and then price the rest term using the worst possible weights, given today's market values