Methods for "prompt month equivalent" exposure in commodities forwards/futures markets

Question

It is common in commodities markets to hold many positions, both long and short, across a range of contract months beginning in the prompt month (today, September) to five or more years out. In general the prompt month exhibits the most volatility, and far out months exhibit the least (among the same exact products)

This makes total notional 'position' for a product quite misleading. For example if today I purchase 1 September 2014 contract, and sell 1 September 2019 contract, my net notional position is 0, suggesting the portfolio has no risk. In reality if the prompt month appreciates 10%, the 2019 contract will appreciate maybe 1%, and you have realized a significant gain.

Standard VAR calculation process captures and handles this well, but I want to be able to measure my true spot price exposure for a product class for purposes of separating position limits from VAR limits.

So, the question,

What is the industry standard model for condensing a strip of forward contracts into a single exposure number "FME" such that it is reasonable to approximate PNL by taking FME*spot price. (presume you are given a corr/cov matrix)?

The most relevant article I can find is here. My calculations from that paper seem somewhat accurate and it covers the subject quite well subjectively, but it has no citations and I doubt my peers would accept it as a source for policy.

http://kiodex.com/our_library/FME_Whitepaper.pdf

Answer 1

I wouldn't say that there is a single industry standard. Also, I'm not sure you should try expressing all risks in a single number, or at least be aware of it's deficiencies.

If you're interested in VaR, you could consider doing historical VaR for example (be sure to correct for seasonality effects and the curve rolling down with the passage of time).

One approach that I personally like, is a PCA based approach, similar to what people do with interest rates. On the basis of deseasoned, roll-corrected returns, you can perform a PCA analysis; depending on the commodity, you typically find 1-3 factors explaining 95+% of the variance. You could determine the exposure of each future/forward you have in position to those factors and sum those?

Yet another approach could be assuming explicit multi-factor dynamics for the curve, calibrate your model and based on that come to a risk number.