[R-SIG-Finance] Does Nelson/Siegel analagous model exist parsimoniously to paramaterize an FX volatility surface?

[thomas.browne at mac.com]

Hello,

I have successfully implemented modelling of the South African and other yield curves in R using the YieldCurve package suggested in this forum, which allow me to describe most states of the yield curve using 4 parameters. The advantage is that these four parameters "make sense" for most traders in yield curves, since they correspond to the intuitive factors of level, slope, "hump", and hump peak location. The further advantages of this paramaterized model is that it doesn't force orthogonality of the factors (as PCA would do, for example), allowing me to apply regressions to the various factors and therefore see when, for example, slope is out of sync with level or hump (with all the caveats of historic data etc etc). Finally Nelson/Siegel has the excellent property of being stable at longer maturities. I am finding this very useful for strategy selection. 

Now I am wondering if an analogous model exists for FX (or other) volatility surfaces. Obviously in this case it would be a higher order model in two dimensions, but I am hoping that such a model might exist which allows me to paramaterize the vol surface. Perhaps somebody has experience modeling the surface using an extension of a Nelson/Siegel model, or another similar approach? 

The factors in this case would obviously again include level, but we would have skew, smile, and term structure of those variables as well, perhaps described in less than 3 additional terms. Again I wish to stay away from PCA as orthogonality is not always desired when comparing one factor against another, especially when the intuition of human traders needs to be considered. 

Ideally I would like to avoid polynomial-based approaches because NS has the superior quality of being stable at longer maturities due to the decaying nature of its terms. I can see how polynomial terms might be appropriate for the skew and smile factors, but I am sceptical on the term structure dimension, for similar reasons as as in interest rates. Anyway just throwing a few questions into the mix.  

Any pointers or help on this is much appreciated, and I will feedback as I progress. Please note that I am not looking for an arbitrage-free model as this is not being used for pricing derivatives, but for medium term (1m to 3m) strategy selection by comparing current surfaces parameters to their history, and to their peers. Thus it would be preferable for the terms to correspond to intuitive factors, as per Nelson Siegel. 

Many thanks. 

Tom