Ventana software support forum

Hello,
I saw in some of the references that SD is not very useful for predicting, but I do not know the reason. I wonder if anybody can help me

Totally wrong.

There is a kernel of truth, which is that forecasting is often not a very high-leverage activity. The "if we know what will happen we'll know what do do" mindset doesn't work in systems with feedback, where actions influence the outcome.

However, whether you want to forecast well, or just make good decisions, either way you need a good model. An SD model is often the best you can do.

thanks, but I did not get the point, especially that the above sentence is some how against the lower one.
Regards

I shouldn't have said that there was a kernel of truth, because the original statement does not contain one. "SD is not very useful for predicting" is a wholly wrong statement.

Whether you want to predict the effect of a policy change, or forecast the state of the system at some point in the future, you need a good model.
http://metasd.com/2011/07/modeling-is-not-optional/
A good model should account for dynamics, behavior, diversity and available data. Often the best model will be an SD model. Sometimes it will be something else - a discrete event or agent model, or something more purely statistical.

When people say "SD is not very useful for predicting," they're generally thinking one of two things:
* One is the idea that forecasting is often a useless activity, and therefore a poor objective for an SD model (but also for any method).
* Another is the idea that SD can't handle data. That's been an obsolete perspective for decades - certainly since the publication of Vensim.

By the way, what references said this?

thanks a million
Ive got it now