I'm wondering what experience other people have had with modelling vol and correlation in commodity markets. I haven't uncovered a concensus from the literature even though the problem looks somewhat similar to interest rates which is well developed. To be clear, I am thinking of no skew for the moment. There seem to be a number of different streams for modelling commodity vol and correlation, some which are studied by multiple authors. I would have thought that something like Rebonato's approach would be the standard, although he doesn't allow for a term structure of correlation in the papers I've seen. The Clewlow Strickland approach could be made seasonal. FEA's seasonal PCA is one of the few tractable papers I've seen which looks plausible, although it won't calibrate to implied correlations without further work. There's a paper by EPRI paper has an interesting analysis for a two factor analytic model . Blix, amongst others, have used fourier series to model the seaosnality, but I suspect the parameter estimation will be complicated, and a market model approach is preferable when correlation and volaitltiy are of equal importance.EPRI: http://www.brattle.com/Publications/Rep ... D=1003Blix: http://www2.hhs.se/efi/summary/657.htmFEA: www.fea.com/resources/a_principal_compo ... alysis.pdf

hi easy, which other texts/authors have best supported your thinking on this matter?

Hi Edouard, The FEA seasonal PCA comes closest to what I have used. One concern I have is with the reliability of the resulting components when you are not applying a functional form or using a filter on the returns before calcualting the covariance matrices. One reason I like it is that it doesn't force a view on how the market is behaving (which is a bit contradictory I know). Most people I have worked with find the approach appealing, although this has been without trying for model skew.

Regarding vol, consider CEV or a modified CEV with mean reversion. Prices up, vol up with commodities.For seasonality consider Geman & Borovkova to identify and remove the seasonality embedded in a futures curve. I'd stay clear of fourier or trigonometric functions to model seasonality. Seasonality is often more like a sawtooth.