Kernel density estimation

[luisangelguzmang]

Hi, it is possible to create a Kernel density estimation in vensim DSS?

[tomfid]

The short answer is yes.

However, there aren't any tools for this specific purpose, so it's up to you to develop and equation structure and optimization control file that will get the job done.

If you say a bit more about what you're trying to do, we may be able to be more helpful.

[luisangelguzmang]

Thanks for your help. Well, I need to create a kernel density function from a income matrix (each row is a incomne value, and the column is the year). So can you help me?

[tomfid]

Are you trying to smooth across time, or income values, or both?

If this is all you're doing, you'd be better off using a stats package. But if this is part of a dynamic model, we can probably come up with a strategy that works.

Tom

[luisangelguzmang]

I need to calculate the density functions by income across time (one function for each period of time), and yes, it is a part of a dynamic model.
thanks again....

[tomfid]

OK - I think this is actually fairly easy. I'll post an example.

One more question though: once you have the kernel-smoothed density, what will you do with it? For example, do you want to pick arbitrary points from it, or generate incomes at uniformly-spaced levels, essentially using the pdf as an interpolator? Or ... ?

[tomfid]

Here's an example:

kernel density.mdl

(3.7 KiB) Downloaded 220 times

I haven't double checked to be sure that the resulting pdf integrates to 1, but it should be OK.

[luisangelguzmang]

Thank you. I will check the example. Basically, after kernel-smoothed density, I need to pick a single level of income (each period of time), and then calculate how many people is below (or above) to this level of income....

[tomfid]

OK. I think my example should work for you.