Could someone give me a good example of fixed-income trades that specifically need stoch vol ? (I can think of a few but I just want to check myself here...)

option on a vol bond

Why don't you list the ones you've thought of....Then I'll add any I can think of that you're missing.

Well, I could only think of options on vol bonds indeed, and variations of that. For vol bonds, you don't need a stoch vol model neither, because you just depend on the outturn joint distribution fothw to fixings/rates, which you can fudge with extrageneous correlations.Could you think of anything else ?

I think it depends what you mean by "need" stoch vol model.You are right with the option on vol bond.However, almost all structures would benefir from an "accurate" stoch vol model.i.e. anything that has heavily OTM or ATM strikes. You can price these without stoch vol, but more often than not, people will be queing up to trade with you.

How about out-of-the-money CMS cap?

bsycheng: I suppose an objective measure of stoch vol model's requirement is done by comparing volgamma and/or vanna (and other second order cross gamma risks) with the vega of the trade.The problem of course is numerical. For those trades where we are not sure, we usually need a term-structure model to calculate their PV in the first place.If the only model available for the type of trade we're looking at can be priced on Monte-Carlo only, then we hit the problem of having to calculate second order risks in Monte-Carlo, which, combined with numerical noises, could yield false alarms as well or hide real risk noises. So this is why I asked the question in the first place. In other words, apart from option on vol bonds, has anyone seen some liquidity/market for fixed income trades that really need pricing on a stoch vol model ?

My firm (and me too) may not be as sophisticated as Piterbarg's in terms of modelling, but from a model user's point of view, I try to avoid stoch vol as much as possible, even when pricing would clearly benefit from it. The problem is that calibration is not only a pain in the ass, it is also less stable, and what is more important: the management of the deal once it is in your book becomes burdensome as you need to check the calibration regularly in order to trust the hedge ratios. I rather take a simple model, add a hefty margin...and if in doubt have one of the quant's double check price and risk vs a more sophisticated tool.Before anyone starts making fun of this approach: the problem is not only to choose the right model for the right product, the model itself needs to be implemented such that it can actually be used in the most pain-free manner. Imagine driving an SL500 without any of the amenities built in by mercedes (ESP, power steering, A/C, navigation etc etc)...a 1990 Bronco would be just as good.

Your view is interesting because it suggests that you can actually do business without a stoch vol model at all ?I suppose you're taking alot of reserve at trade conception, but you could face (at least) two fundamental problems. One is that having to take so much reserve in the first place could seriously limit the development of the business in that type of product. The second problem of course is that your book could be leaking P&L, and could be leaking much more than the reserve you had in the first place.So is your experience so far that these two issues, though correct in principle, do not pose too much problem in practice ?