I am very new to the field and I was trying to understand the differences between approaches for generating scenarios in SP. From what I have seen, people have mostly focused on moment matching methods followed by some discretization to reduce the scenario space. However, how does this compare to a timeseries approach where one first makes a prediction for each action/solution in the future and then optimizes based on this (in an MPC way of solving the problem)? I am a practitioner so I would be very grateful if someone can give some practical advice on this. asked 25 Nov '15, 05:22 lstavr 
I am not familiar with timeseries approach. ScenarioGeneration sound like SAA(Sample Average Approximation) Method. Stochastic programming problems usually have expectation of some function with random variables, in its objective function. To get the exact solution the expectation should be exact. However, many problem have complicated distribution functions or too many (even infinite) possible outcomes. In order to solve these intractable problems, some scenarios can be generated (of cause according to the distributions) and the average can be used rather than exact expectation. Obviously, it is approximation method. Google 'Sample Average Approximation' and you can find tons of documents. answered 29 Jan '16, 23:02 ksphil Thank you, I was interested in multistage stochastic programming problems, where at each stage a number of possible actions are possible so some measure of uncertainty is required for each action. I will have a look at SAA!
(31 Jan '16, 18:07)
lstavr
I understand. SAA is not a tool for multistage problems. However, it can be applicable to multistage problems also. It might be even more powerful in multistage problems, in some sense.
(31 Jan '16, 18:14)
ksphil
