Wanted to see if anyone has further insight into this. Problem: Simulating a portfolio of large number of commodity positions (e.g. NG). Lets say we have a portfolio of 30,000 small positions on NG Nymex contracts from today till next year. (i.e. until Sep 09 contract for example). It is relatively easy to do a MC simulation of these contracts and evaluate the positions if it were not for the fact that there are 30K positions and most of them are daily options (though settled on the Nymex monthly contract)A possible solution is: Since all the positions are based on monthly Nymex contracts, simulate the Nymex contracts, then use DG approximation to evaluate the portfolio. But since the time horizon is relatively large, the Dg approximation does not work very well in the tails of the distribution. Infact it blows up under certain scenarios. Can anyone think of another way to simulate the portfolio?

1. Simulation for what purpose? (VAR, valuation, etc)2. What's a "DG approximation"?regards,

1. VaR and other risk metrics2. Delta gamma approximationThere are few tricks i can think of. One is to find Delta gamma at different percentiles and use those in DG approximation. but wanted to see if anyone else has any other ideas.Thanks

I see. Well, not my market -- nevertheless, I'm not sure I see why having 30K positions matters and why you need this DG approximation.For example, suppose I had 30K SPX futures positions, all maturing by July 29, 09 or earlier.For VAR, I would simply set up some decent wide-tailed distribution for thedaily SPX returns and draw one return per day for 252 days. Repeating say 10^5 times is a good simulation. The simulation run simultaneously values *all* 30K positions with no DG approximation, right? Now for SPX you might have a stationary distribution and likely in your market thereis seasonality and maybe trickier to convert underlying NG returns to futures returns, etc. But once you adjust your daily distribution to account for the time of year (and these otherfactors special to your market), it still seems like one simulation run handles the whole batch. What am I missing?

Complication comes from the fact that all the positions are not same contract. most of them are daily options which expire on different days, some of them are monthly contracts again expiring at different months. If we do a daily simulation, now i need to get correlaton among these contracts etc...

Why? Let's say you do the daily simulation I suggested to Jul 29, 2009 andone of your contracts expires on Sept 19, 2008. At the start of the simulation,I say OptVal091908 = 0// For each day of my simulation I know the corresponding date and futures price F, so I say ...If(date == 091908) OptVal091908 += Max(F - K, 0)// At the end of the simulation, I say OptVal091908 = OptVal091908/NSIM// Same for every other contract.