Hi,suppose I have a position in a commodity like Brent oil which consists of a variety of different futures on that commodity with various expiry dates. They are all on the same commodity. What is the best way to calculate my value at risk for my total position?-CM

One of the biggest factors that make this different from other e.g. equities is the samuelson effect (volatility of a future increases as it comes closer to maturity). Depending on what commodity you're talking about, seasonality is another one.Not making any claims about this being the best way, but if I would have to do it, I would calculate VaR in 3 steps:1. If needed (for Brent it isn't) remove any seasonality effects.2. Calculate a price history for a number of dummy futures with fixed maturities, 1m,2m, 3m, etc. From this, construct timeseries of price moves for the various dummy futures.3. For each future in your position, determine the time to maturity and interpolate the abovementioned timeseries to get price moves for the specific future.I'm sure others will chime in with improvements

We used to do it like BramJ said, there are probably better ways but the backtests were goodAlso, there is a big difference between types of commodity - brent is a homogenous product with regular expiries, but in a lot of ag products for example different expiry months will correspond to different crop years, so each future would have to be allocated to old crop or new crop and you would need two series for the same product because there are effectively two different underlying assets, even though they are all notionally wheat or soybeans or whatever

QuoteOriginally posted by: xpatagonWe used to do it like BramJ said, there are probably better ways but the backtests were goodAlso, there is a big difference between types of commodity - brent is a homogenous product with regular expiries, but in a lot of ag products for example different expiry months will correspond to different crop years, so each future would have to be allocated to old crop or new crop and you would need two series for the same product because there are effectively two different underlying assets, even though they are all notionally wheat or soybeans or whateverWhat kind of time series (predictive?) model was used?

QuoteOriginally posted by: xpatagonbut in a lot of ag products for example different expiry months will correspond to different crop years, so each future would have to be allocated to old crop or new crop and you would need two series for the same product because there are effectively two different underlying assets, even though they are all notionally wheat or soybeans or whateverthat's interesting.thus any ag futures series necessarily involves a single year crop as the behavior of different each year crop is quite unique ?thus creating some long-term ag futures series to build an ag futures forecast models is useless?(I'm not being cynical at all. I have been wondering for several years now whether ag price forecast models are worth at all)

Its not so much that each year is unique in itself, but that according to the month of the year the deliverable product will be different. For example, if your crop is harvested once a year between october and december then delivery months october to january probably correspond to product that is delivered ex farm, and other months product that is ex warehouse, so if I buy a future in May with august delivery then my underlying product will be warehoused grain from last harvest, but if I buy for november delivery the underlying product will be product from the upcoming harvest. In terms of futures this means that futures deliverable up until october will correspond to last years grain, and futures deliverable after october to next years grainDepending on anticipated supply and demand, the relative quality of the two harvests and other factors there may be strong market preferences for one or the other.In reality it is a lot more complicated than that, with global suppliers, crops with two harvests per year and (I suspect) feedback mechanisms from crops that are priced basis the future, but the basic concept is that depending on the time of year the nature of the physical supply changes as well as the quantity supplied and this feeds back into the futures.I am not really convinced about absolute price forecasting, but I have known people who were very good at local market volume forecasting (eg Brazilian soybeans, Australian wheat etc), which obviously feeds through into better trading.The ag products I have worked with were all non-US products, so the basis was generally more important than the future - depending on the relative importance of the US underlying vs the real global market it may make more sense to model an overall price level and major supplier basis then back a futures price out of that. Just a thought though, I havent looked into the mechanics of it.

Hi,Some markets would have a seasonality in the volatility, e.g. the power market, but this should be linked to the traded products, no? The Samuelson effect explains why the volatility is declining in time to expiry, but there is more structure in many markets. Extracting the seasonality might be difficult...Typically, one would collect historical data on month-ahead, two-months ahead, ... and build a long history of forward curves. You could apply PCA to decompose, and see how many factors would be needed to describe the variance. VaR could then be calculated (and then incorporating correlations) based on position size and expected volatilities. Googling VaR and PCA will provide you with many illuminating examples. S