I want to know how to avoid winner-take-all. The objective function of the TIMES model is cost optimization, which can impose market penetration or technology share constraints. However, for multiple technologies, if every process is set (such as the transportation sector), there may be some questions such as 1) Not every process can be constrained. And if we add too many constraints, it is the intervention to the model? In this case, imposing economic policies such as carbon tax may not work, because the cost caused by the imposed carbon tax is not high enough or due to the constraint. And if no constraint is set, the model may select the process according to their costs and resource endowment, and the use of suboptimal option only after the lowest cost is used up. How do you solve this problem in the modeling process? Whether the TIMES model can be linked with a discrete selection model similar to the GCAM model, and the market share can be calculated based on the cost.
>Whether the TIMES model can be linked with a discrete selection model similar to the GCAM model, and the market share can be calculated based on the cost.
Yes, sure it can. Some TIMES models and the Markal SAGE algorithm implement market share algorithms comparable to the GCAM modified logit discrete choice model, see below:
As you can see, both the TIMES Ireland model and the Markal SAGE time-stepped algorithm implement market share algorithms that are resembling the GCAM Modified Logit formulation. The CIMS formulation adopted by TIMES Ireland differs from it by employing intangible costs instead of preference/calibration weights.
Years ago, it was suggested that the automated Markal SAGE market share algorithm be implemented also in TIMES, but thus far it has not been implemented. However, should there be general interest in such, it could be again considered, provided that the design for it is first reviewed for TIMES and improved where necessary.
In addition, I think one might also consider utilizing soft market share constraints instead of rigid ones, such that increasing/decreasing the market shares from the predefined bounds would be possible with some additional cost, or cost curves. For example, a high carbon tax would thereby be able to realize some additional shifts in technology market shares, beyond the default bounds.
Thanks for your reply. That is, we can fix or UP（LO） to limit the share of process based on their costs, is it right? And when levying the carbon tax, the cost of process which emitting carbon emission will be changed, and then we should fix the share of process again based on their cost? Here is the question, when setting the carbon emission, how to decide the share?
Anna
(13-12-2022, 12:24 AM)Antti-LThanks for your reply. That is, we can fix or UP（LO） to limit the share of process based on their costs, is it right? And when levying the carbon tax, the cost of process which emitting carbon emission will be changed, and then we should fix the share of process again based on their cost? Here is the question, when setting the carbon emission, how to decide the share?Anna Wrote: >Whether the TIMES model can be linked with a discrete selection model similar to the GCAM model, and the market share can be calculated based on the cost.
Yes, sure it can. Some TIMES models and the Markal SAGE algorithm implement market share algorithms comparable to the GCAM modified logit discrete choice model, see below:
As you can see, both the TIMES Ireland model and the Markal SAGE time-stepped algorithm implement market share algorithms that are resembling the GCAM Modified Logit formulation. The CIMS formulation adopted by TIMES Ireland differs from it by employing intangible costs instead of preference/calibration weights.
Years ago, it was suggested that the automated Markal SAGE market share algorithm be implemented also in TIMES, but thus far it has not been implemented. However, should there be general interest in such, it could be again considered, provided that the design for it is first reviewed for TIMES and improved where necessary.
In addition, I think one might also consider utilizing soft market share constraints instead of rigid ones, such that increasing/decreasing the market shares from the predefined bounds would be possible with some additional cost, or cost curves. For example, a high carbon tax would thereby be able to realize some additional shifts in technology market shares, beyond the default bounds.
If you use one of such Logit approaches, it would be the heterogeneity (γ) and preference weight (α) parameters that you should decide, where the preference weights are mainly for calibration, and heterogeneity describes the sensitivity of choice to the cost differences.
I would like to ask whether there has been developed any documentation using previously mentioned Logit approach in TIMES.
My goal is to implement Logit approach to our currently used Czech TIMES model.
If not, could you recommend what steps need to be undertaken (code adjustments) for ensuring proper function of TIMES including Logit.
Thank you for your reply!
We have tried to implement changes according to the shared documentation but we have not succeed yet.
Could you please instruct us where into the GAMs code should the line of code: $SET ECB YES be added.
Further, we would like to implement market share allocation on transportation sector.
To do so we have tried to adjust demo model number 12.
Could you possibly send an instruction how tables for COM_MSHGV and NCAP_MSPRF should be implemented into the SuppXLS folder and what sort of an index should be at the top of the table.
23-10-2023, 02:47 PM (This post was last modified: 23-10-2023, 03:03 PM by Antti-L.)
No GAMS code changes are required to be done by the users. The switch can be specified in SysSettings any regular scenario file, see: Modifying RUN files
Only COM_MSHGV is required for using ECB. As described in the documentation, NCAP_MSPRF is optional, and should only be used when preference weights or intangible costs are also modelled, or if the the default penalty costs are to be modified. Both parameters can be defined in any scenario file. The indexes of the parameters are also described in the documentation:
>what sort of an index should be at the top of the table
As mentioned above, you should specify the Region, Year, and Commodity indexes for COM_MSHGV. In a VEDA TFM_INS table, for example, you can put Region indexes at the top of the value columns or use an AllRegions value column, and for the Year indexes, you can use the Year column, and for the Commodity indexes, you can use the CSet_CN column. You can find many examples of using TFM_INS tables in the Demo models.
23-10-2023, 02:56 PM (This post was last modified: 23-10-2023, 02:57 PM by AKanudia.)
Thanks, Antti. Just one clarification - Tags like "RFCmd_Flags" can be used in regular scenario files too. So, I would set up a scenario file where you declare "$SET ECB YES" via this tag, and declare the _MS attributes. One could also relax other FLO_SHAR/FLO_MARK that may have been used to guide the model before using this approach - in the same scenario file.
This would enable switching the logit approach on/off at run time.
09-11-2023, 07:23 PM (This post was last modified: 09-11-2023, 07:24 PM by Lukas255.)
Dear all,
further I would like to ask how intangible cost influence the objective function?
We simulated 2 different scenarios which generated results leading to opposite conclusions.
We simulated demo_model_12 with the intangible factor set to 0.5 and 1 according to the documentation. In the next step we simulated intangible costs for only one process DLS WITHOUT possibility of substitutes which are available in demo_model_12 such as (ELC, BIO, ...).
Although, first application works as expected, the second application does not work as expected. The data show lower costs for DLS from original model and higher costs for both models with additional intangible costs set to 0.5 and 1.
Secondly, I would like to ask about possibility of setting endogenously parameter alpha dependently on predicted future demand, fuel prices and so on. Is this also possible? Are you planning such a feature?
>further I would like to ask how intangible cost influence the objective function?
In the formulation, they don't. They are not considered real costs, but only perceived costs affecting the economic choice behaviour.
>The data show lower costs for DLS from original model and higher costs for both models with additional intangible costs set to 0.5 and 1.
Intangible costs represent an additional cost component in the logit choice formulation, and thus may smooth out the true cost differences, and therefore lead to higher overall costs (further away from the optimal solution). However, I am not able to comment on your findings, because e.g. I don't know what you mean by "data show lower costs for DLS". What data shows which costs for DLS?
>Secondly, I would like to ask about possibility of setting endogenously parameter alpha dependently on predicted future demand, fuel prices and so on. Is this also possible? Are you planning such a feature?
If you wish to suggest such an enhancement, could you please provide the mathematical formulation for those proposed endogenously derived alpha parameters?
(09-11-2023, 07:46 PM)Antti-L Wrote: >further I would like to ask how intangible cost influence the objective function?
In the formulation, they don't. They are not considered real costs, but only perceived costs affecting the economic choice behaviour.
>The data show lower costs for DLS from original model and higher costs for both models with additional intangible costs set to 0.5 and 1.
Intangible costs represent an additional cost component in the logit choice formulation, and thus may smooth out the true cost differences, and therefore lead to higher overall costs (further away from the optimal solution). However, I am not able to comment on your findings, because e.g. I don't know what you mean by "data show lower costs for DLS". What data shows which costs for DLS?
In the attachment I send you adjusted files for demo model 12.
We have erased all technologies except TCANDSL1 and TPUNDSL1.
In the scenarios we set intangible parameters to 1 and 0.5.
The unexpected result was in investment costs for new technology where as base was a case model 12 scenario (predefined) and as observed were set cases ITC1, ITC05 where Scen_CH2_5_LO_1 and Scen_CH2_5_LO_0.5 were added accordingly to base case model 12.
However investment costs for the new technology TCANDSL1 for cases ITC1 and ITC05 were higher then base case but both had same value.
