Ventana software support forum

Hi all,
I have another question about the DICE model taken from Tom Fiddaman's web site.

I noticed that the optimal Nordhaus paths are included in the model as "external" variables. The question is : why can't we run the optimization directly in VENSIM? I think we can run the model asking VENSIM to optimize the Welfare (or Utility ) function. Am I correct or did I miss something ?

Thanks in advance
G

I try to ask the question in another way.

Can I replicate the DICE model in VENSIM letting VENISM do the optimizations as it's done in the GAMS language instead of importing the optimal path from the results obtained by GAMS into a VENSIM Variable?


Thanks a lot in advance
G

Hi Duilio

Can you post the model in question?
Regards.
JJ

Sure,
here's the Model .

I registered myself as Duilio but my name is Giovanni

Hi Giovanni

As I see that you are on line, did you ask Tom Fiddaman any
question about the model?
Does Gam makes optimization with each time steps or along the whole time period?
Regards.
JJ

Hi

It is impossible to run the model, because the data files are not included.
Regards.
JJ

My apologizes. I've attached the data file.

Generally speaking my question is on policy optimization.
Let's imagine I want to find the consumption path that maximize utility in a macroeconomic model.

Can you give me some details on how VENSIM works?
I've read the online help but It's still not 100% clear.

Thanks a lot, you're one of the pillars of this Forum!!!
G

Hi Giovanni,

On the DICE model Tom Fiddaman can give definitive answers, I would recommend you send him an email.

To do a path optimization in Vensim you need to break the path up into a series of intervals. For example you might use

tperiod : (t1-t10)
investement = VECTOR LOOKUP(investment profile[t1],INITIAL TIME,FINAL TIME,0)
investment profile[tperiod] = 1,2,3,4,5,6,7,8,9,10

the optimize over investment profile.

Hope that is helpful.

Hi Giovanni

I do not want to dodge the question. But Tom is as Bob says the best able to answer the question.
About optimization, finding directly the path optimizing different periods independently works well in a closed system but if there are exogenous inputs it becomes more difficult unless you can model the inputs with a mix of random functions and try to optimize the average of the pay off to optimize. Another solution is to find a rule that calculates the investment of the next period based of the results of the preceding period and optimize that rule. The problem is to find the adequate rule.
The advantage of rule optimization is that it requires few parameters. In the case of a great number of periods, the direct optimization may become problematic for a question of time.
If there are many periods (say 40) and you need to run parallel subscripted simulations (say 100 as I generally do) the model may take a very long time to optimize, without being very sure of having the best solutions.
Another difference is that direct optimization does not add any more feed back loop, rule optimization adds a feed back loop.
Regards.
JJ

Guys,
thanks a lot for your inputs.

Cheers
G