GEV draws for the MDCNEV
Posted: 17 Sep 2023, 09:10
Hi,
this is a bit of a mix of questions regarding the generation of GEV draws for forecasting using an MDCNEV model and application of the MDCNEV in general. I have been trying to wrap my head around how to generate GEV draws, as I ultimately want to apply an MDCNEV model within an agent-based model.
As stated in the manual, I understand that the forecasting in apollo is based on the approximation procedure proposed in https://doi.org/10.1016/j.tra.2017.07.003
I have a couple of questions on that procedure:
1) How exactly is the correlation matrix Omega defined. Is the dimension given by the number of nests, i.e. SxS? And the elements given by the approximated correlations, e.g. in a 2-Nest case:
(1-Theta_1^2) 0
0 (1-Theta_2^2)
2) Where does the "12" in the "R x 12 matrix" in step 2 come from. Is this because you have 12 alternatives?
3) If I look "under the hood" of the apollo code, would this be implemented exactly this way or did you have to make some adjustment for it to fit into the "apollo universe"?
And finally (a much more general question for using any MDCEV or MDCNEV model in an agent-based model):
4) Do you have any pointers on how best to determine a specific consumption pattern for each individual agent? I know that Bhat has proposed some procedures (https://doi.org/10.1016/j.trb.2018.02.011) but I was wondering if you have any experience (or anyone in this forum for that matter).
Thanks!
Anna
(Sorry about the formatting; I guess a final question: is there a way to integrat Latex formulas in this forum?)
this is a bit of a mix of questions regarding the generation of GEV draws for forecasting using an MDCNEV model and application of the MDCNEV in general. I have been trying to wrap my head around how to generate GEV draws, as I ultimately want to apply an MDCNEV model within an agent-based model.
As stated in the manual, I understand that the forecasting in apollo is based on the approximation procedure proposed in https://doi.org/10.1016/j.tra.2017.07.003
I have a couple of questions on that procedure:
1) How exactly is the correlation matrix Omega defined. Is the dimension given by the number of nests, i.e. SxS? And the elements given by the approximated correlations, e.g. in a 2-Nest case:
(1-Theta_1^2) 0
0 (1-Theta_2^2)
2) Where does the "12" in the "R x 12 matrix" in step 2 come from. Is this because you have 12 alternatives?
3) If I look "under the hood" of the apollo code, would this be implemented exactly this way or did you have to make some adjustment for it to fit into the "apollo universe"?
And finally (a much more general question for using any MDCEV or MDCNEV model in an agent-based model):
4) Do you have any pointers on how best to determine a specific consumption pattern for each individual agent? I know that Bhat has proposed some procedures (https://doi.org/10.1016/j.trb.2018.02.011) but I was wondering if you have any experience (or anyone in this forum for that matter).
Thanks!
Anna
(Sorry about the formatting; I guess a final question: is there a way to integrat Latex formulas in this forum?)