I seem to miss a crucial point of yours. You said in an earlier post:Quote That being said, your answer will also be MASSIVELY model-dependent. The kind of seasonality and jumps in gas prices that you think there are, and the nature of your assumptions about summer/winter price spreads is going to have a huge effect on your optimal injection/withdrawal strategy and, therefore, your option-equivalent price. Modeling gas storage optionality is what we call a "hard" problem: choosing which measure to use is not the biggest problem you face Hence, on the one hand side you say that the calculated value will depend (massively) on whether the gas-price process will have jumps, the kind of seasonality, etc, i.e., on the specification of the underlying process (I totally agree to that point and saw this happen often enough). On the other hand side you state that I should not worry too much which measure/model to use. This seems contradictory to me. Let me also state that once I have a stochastic model, I feel quite comfortable in solving the resulting stochastic optimization problem used for storage pricing (that is my main area of expertise, not finance, as you might have noticed).(ad 2): I can see that stochastic volatility might be important in gas markets. I am also not too happy with modelling the spot price - I would rather go for a model of future prices directly (instead of calculating them from spot prices). Hence, the Schwartz&Smith paper was just intended as an example of a model that simultaneously yields a physical and a risk free measure. My preferred option would be a multi-factor GBM process for the forward curve. Of course, for that the question on how to calibrate the convenience yield and the market price of risk remains to be answered (I am thankful for any hints/suggestions).(ad 4): To my understanding even with a constant convenience yield (i.e., only one parameter) the price of any specific future would have a tendency to rise as it 'moves towards delivery'. This remark was an attempt to answer your question why future prices should have a trend.

Oh, Lord, no. I'm certainly not saying the model doesn't matter. The model matters a lot. All I meant is that your other modeling choices are going to dominate choosing whether to use "r - q" or "r + lambda - q" in your drift specification, for example. You will need the "lambda" if you are estimating the model, but you should be careful about believing too much in its value.I actually think it's generally better *not* to model the actual HH spot price in practice, though this rather depends on whether you're involved directly in physical markets or not. Then you are left with either modeling the futures prices directly, or modeling a "virtual" spot price which is a hidden state that you only see via the futures prices. This tends to cut down your degrees of freedom and simplify the model a little. Not modeling the spot price directly will also mean it will be easier for you to price options on futures, which you might want to use to calibrate your extrinsic storage price. Of course, for gas, you still have to think about how to get the summer and winter prices to decorrelate usefully.Have a look at Gibson & Schwartz 1990 (stochastic convenience yields), Cortazar & Schwartz 2002 (an augmented 3-factor version of that). For a quite useful approach to seasonality in a stochastic vol setting, you might check out Richter and Sorenson 2003. It's about soybeans, but could be useful, as it lets you get convenience-yield seasonality in a fairly parsimonious way (realistic is another matter, of course )The futures price should still have zero risk-neutral drift, shouldn't it, because it's an expectation, right? The spot price "drifts towards" the futures price rather than the other way around - the convenience yield affects the cost-of-carry of holding the spot-asset: dS/S = (r-q)dt + s dw and F = exp((r-q)T)S gives dF/F = s dw.