And let me re-iterate: my work in implementing the LMM type models (and variants) did pay for itself many times over, but NOT in princing/managing exotics

QuoteOriginally posted by: PatAnd let me re-iterate: my work in implementing the LMM type models (and variants) did pay for itself many times over, but NOT in princing/managing exoticsSo, I would state the opposite - ONLY full yield models as BGM or HJM can correctly price stuff like callable CMS spread options: 'callable' means you have to calibrate to co-terminal swaptions, if you have CMS spread between 10Y and 2Y rates, you must calibrate to such swaptions as well. No mercy - short rate models just can't do it. Gaussian models misprice long term callable options - so what remains is BGM.

QuoteOriginally posted by: DariuszSo, I would state the opposite - ONLY full yield models as BGM or HJM can correctly price stuff like callable CMS spread options: 'callable' means you have to calibrate to co-terminal swaptions, if you have CMS spread between 10Y and 2Y rates, you must calibrate to such swaptions as well. No mercy - short rate models just can't do it. Gaussian models misprice long term callable options - so what remains is BGM.This statement sounds as if made by a televangelist, and its information content is zero. I suggest that people cool down their emotions so that this interesting discussion can continue in a civilized way.

Is there any empirical paper trying to solve this issue, i.e. comparing a short rate model with LMM, HJM in pricing and hedging an (or perhaps a portfolio of) exotic deal(s)? Im sure banks has done this but I don't think it's in there interest to make in common knowledge.If one were about to perform a test, what would one look at? Which models? Deals? Times?

A. Unfortunately, how well a model is implemented is often more important than which modelB. If I'm pricing a spread option (say, 10y-2y), believe me I have a factor dedicated to handling the spread ... treating the spread as the difference betwee two comparably sized factors would make me nervous since small errors in either factor can alter the spread substantially

QuoteOriginally posted by: PatA. Unfortunately, how well a model is implemented is often more important than which modelI agree. And the problem with BGM is that there are too many publication some of them misleading.QuoteOriginally posted by: PatB. If I'm pricing a spread option (say, 10y-2y), believe me I have a factor dedicated to handling the spread ... treating the spread as the difference betwee two comparably sized factors would make me nervous since small errors in either factor can alter the spread substantially I used to construct many single factor models to capture specific underlyings (as CMS spread) one, two years ago. Now, I prefer one universal model, they were not very accurate.PS. I hope Pat, we'll discuss that subject personally in future. Open forum is not good for that purpose, I don't like personal comments from anonymous participants, so try not to post too much.

Dariusx, may all the great men not follow your advices. For the progress of all the mankind.

QuoteOriginally posted by: Stochastic44Dariusx, may all the great men not follow your advices. For the progress of all the mankind.Hmm, you know BGM better than Gatarek, stochastic volatility better than Heston, Gaussian copula better than Li. My respect dude. Are you French?

no