wilmott.com

Hi all,Has anybody worked on the article "A Stochastic Volatility Model for Bermuda Swaptions and Callable CMS Swaps" by Claudio Albanese and Manlio Trovato (28 Nov 2005) ? The attachement to the article is provided below.I'd like to start a discussion on their model and its implementation.Please let me know.

What did they do, what was not done previously? Do you believe in stoch vol short rate models?

I think that one cannot belive in models but can just understand their principles and observe their results.I do not have this stoch vol short-rate one implemented yet. I am working to have it running in C. Short-rate models are known to be old-fashioned compared to forward libor ones. But the stoch vol feature seems to give them a new look adapted to swaption smile and callable swap pricing.One visible advantage is that calculations are deterministic: no random numbers. This leads to stable greeks and reproductible results.The authors plan to extend the current version of their model so as to be able to price CMS spread options.

QuoteOriginally posted by: gpopI think that one cannot belive in models but can just understand their principles and observe their results.I do not have this stoch vol short-rate one implemented yet. I am working to have it running in C. Short-rate models are known to be old-fashioned compared to forward libor ones. But the stoch vol feature seems to give them a new look adapted to swaption smile and callable swap pricing.One visible advantage is that calculations are deterministic: no random numbers. This leads to stable greeks and reproductible results.The authors plan to extend the current version of their model so as to be able to price CMS spread options.Are you able to justify your statements? You know - creating ad hoc new interest rate models was popular in mid nineties and almost all of them are dead now, even a very interesting model by Flesaker-Hughston. Libor BGM and its mutations like swap BGM, will remain the pricing standard forever. I suggest credit risk if you like ad hoc models - it's exactly what happens in the CDO reserach now. You create a new copula, prove it fits better than the previous one for a chosen iTraxx or CDX and have a PhD, publication o quant position, whatever you want.

Of course. I already have one quant working on Libor MM, which will be extended to Swap MM. That they will last "forever" seems quite presomptuous. They are Monte-Carlo-based and I am interested in comparing their behaviour with deterministic formula-based ones.Do you have any comment on the Albanese-Trovato article ?

Hi gpop,did you make some progress in evaluating stochastic vol short models yet?Would be interesting to hear if these non-MC implementations worth consideration.

Hello everyone, I think the point Manlio and I want to make is not that of proposing yet another short rate model. As a matter of fact we wrote several papers recently, each one with a new model specification. Our main objective is rather to develop a viable numerical framework for models specified semi-parametrically without any assumptions of analytic solvability. I have been posting a number of papers over the years, each one with an improvement on the numerical side of things. At this point, I believe the modeling framework we arrived at is viable for industry strength implementations and is very competitive for exotics like callable CMS spread range accruals, TARNs or snowballs. In my last paper I have benchmarks (see http://www.level3finance.com/irmodel.pdf). Refer to this one for the best methods I advocate at this point in time.As lattice algorithms, these methods are characterized by hourly time steps and a new way of pricing path dependents. The very same models can also be reinterpreted and execute as Montecarlo algorithms, whereby one simply uses the kernel to generate scenarios under the risk neutral measure and there is no need to impose drift restrictions. For correlations and hybrids, one can then use dynamic conditioning as in my CDO papers, or Montecarlo simulations. I prefer the former whenever it is applicable as it is noiseless.The main advantage of semi-parametric modeling is that it is very flexible and models can approximate econometric evidence to a good degree. This gives new useful insights and leads to prices and hedge ratios for exotics which may differ from the street values. I now personally think that stochastic monetary policy is a very important concept, more important that stochastic volatility. Meaning, it is important to model directly the stochastic process for the drift of short rates as this controls the volatility and correlation of the long rates in a realistic fashion. Modeling correlations without caring about the drift and simply implying can lead to economic inconsistencies that reflect in awkward dynamic specifications. Stochastic monetary policy can also explain the swaption skew without too much in the way of time dependencies and a number of backbone and convexity features. But hey, if you don't like the economic scenario I am proposing, whatever model you'd like to code would work as efficiently from the numerical viewpoint. The only restriction is that market-models can't be implemented efficiently, only short rate models work well. Otherwise, the numerics are indifferent and decoupled from modeling assumptions. This was not the case with the earlier generation of short rate models and this is the contribution we intended to give.