Optimization Problem

[DominikH]

Hello everybody,

I´m writing a demographics model. I modelled one birth rate and for each age group one death rate.

My problem is that using optimization in Vensim for forecasting, changes are made only to constants (the death rates and the birth rate are constants).

Is there a possibility that i can calibrate my model without writing these constants for each time step? If i´m only changing the inital rates within the optimization process, the outcome is really bad of course.

[bob@vensim.com]

I can't follow your question. Birth and death rates are normally modeled as constants modified by impacts from envionmental, economic and related factors.

If you are trying to compute the per period birth rates based on obseverved population and births that is just a algebraic computation. Useless for forecasting though.

Bob Eberlein

[DominikH]

I tried that as well, e.g.

birth r = pop density * wheight pop density + gdp per head * wheight gdp

but there´s still the problem of optimizing the wheight for all time steps because then the wheights are constant.

Let´s assume optimization would give me:
wheight pop density = 0.01
wheight gdp = 0.00001
These wheight then remain constant over time, maybe fitting the official forecast well in few years but resulting in a huge negative payoff over all years.

From that I went on to optimizing the birth and death rates directly.

DominikH

[bob@vensim.com]

Hi DominikH,

Linear equations are neither robust nor realistic, and tend not to perform well in forecasting for that reason. However, you are attempting to fix this with overparameterized equations, in which nearly every data point has its own constant, and these are completely unusable for forecasting.

The best thing to do is to create a small but robust formualtion fo population changes and then estimate the underlying paramers. The equations in the Limits to Growth model (wrld3-03.vmf) are a good start in this direction.

[DominikH]

Based on your replys and a short look on the world 3 model I conclude that the optimization in Vensim is not suitable for large models and for highly non-linear models.

Would you agree on that?

The next problem is that I don´t have the ressources to build a full economy, environment, etc., so I will try to formalize some lookups (nonlinear) to have my rates fit better into the forecast data.

Thanks a lot

DominikH

[bob@vensim.com]

Hi DominikH,

I am afraid you have totally misunderstood me. However, what you are tryig to do seems like it might be more effectively done using a simple regression model rather than a simulation model. Just be careful that you don't estimate values for more parameters than you have data points.

[Edited on 2006-11-25 by bob@vensim.com]